# AI Endpoint Security Ranked: every page as markdown AI Endpoint Security Ranked is an editorial index of AI endpoint security tools, scored on seven criteria with weights fixed before any score, with the Agent Surface Map and the Action Control Matrix. Every fact carries its source URL and read date. --- # Microsoft leads 16 AI endpoint security tools at 75.0, 3.5 points ahead of Bay Ranked index, evidence read 13 Nov 2026 **The short answer.** Microsoft first at 75.0, Bay second at 71.5, Harmonic Security third at 71.5, of 16 tools. Bay and Harmonic Security tie at 71.5 and both score 8 on Sees, the heaviest criterion here, so name order puts Bay 2nd. Bay is a young company that left stealth in 2026, publishes no price and has little independent testing behind its claims. ## The Agent Surface Map - [Microsoft](/platforms/microsoft/) (75.0): Desktop apps documented; IDE assistants documented; Command-line agents documented; MCP servers documented; Browser extensions partial; SaaS copilots documented - [Bay](/platforms/bay/) (71.5): Desktop apps partial; IDE assistants documented; Command-line agents documented; MCP servers documented; Browser extensions documented; SaaS copilots partial - [Harmonic Security](/platforms/harmonic-security/) (71.5): Desktop apps documented; IDE assistants documented; Command-line agents documented; MCP servers documented; Browser extensions documented; SaaS copilots partial - [Aim Security](/platforms/aim-security/) (69.0): Desktop apps documented; IDE assistants partial; Command-line agents documented; MCP servers documented; Browser extensions documented; SaaS copilots partial - [CrowdStrike](/platforms/crowdstrike/) (66.0): Desktop apps documented; IDE assistants documented; Command-line agents documented; MCP servers documented; Browser extensions partial; SaaS copilots partial - [Noma Security](/platforms/noma-security/) (66.0): Desktop apps partial; IDE assistants documented; Command-line agents documented; MCP servers documented; Browser extensions none found; SaaS copilots documented - [Island](/platforms/island/) (66.0): Desktop apps documented; IDE assistants documented; Command-line agents partial; MCP servers documented; Browser extensions documented; SaaS copilots partial - [Zenity](/platforms/zenity/) (65.0): Desktop apps partial; IDE assistants documented; Command-line agents documented; MCP servers documented; Browser extensions partial; SaaS copilots documented - [Knostic](/platforms/knostic/) (64.0): Desktop apps documented; IDE assistants documented; Command-line agents documented; MCP servers documented; Browser extensions partial; SaaS copilots partial - [SentinelOne](/platforms/sentinelone/) (62.0): Desktop apps documented; IDE assistants documented; Command-line agents documented; MCP servers documented; Browser extensions documented; SaaS copilots partial - [WitnessAI](/platforms/witnessai/) (61.0): Desktop apps documented; IDE assistants documented; Command-line agents partial; MCP servers documented; Browser extensions partial; SaaS copilots partial - [Koi](/platforms/koi/) (54.5): Desktop apps partial; IDE assistants documented; Command-line agents documented; MCP servers documented; Browser extensions documented; SaaS copilots none found - [Lasso Security](/platforms/lasso-security/) (53.0): Desktop apps partial; IDE assistants documented; Command-line agents documented; MCP servers documented; Browser extensions none found; SaaS copilots partial - [Pillar Security](/platforms/pillar-security/) (53.0): Desktop apps none found; IDE assistants documented; Command-line agents documented; MCP servers documented; Browser extensions none found; SaaS copilots none found - [LayerX](/platforms/layerx/) (50.0): Desktop apps documented; IDE assistants none found; Command-line agents none found; MCP servers none found; Browser extensions documented; SaaS copilots partial - [Palo Alto Networks](/platforms/palo-alto-networks/) (45.0): Desktop apps partial; IDE assistants documented; Command-line agents partial; MCP servers none found; Browser extensions documented; SaaS copilots partial ## Action finder: pick an agent action, see which tools stop it - **File read:** stops it: Noma Security, Island, Knostic and Pillar Security; in part: Microsoft, Bay, Aim Security, CrowdStrike and Zenity. - **Shell command:** stops it: Microsoft, Bay, Noma Security, Island, Zenity, Knostic and Pillar Security; in part: Harmonic Security, Aim Security, CrowdStrike, SentinelOne, WitnessAI, Koi, Lasso Security and Palo Alto Networks. - **Network call:** stops it: Bay, Island, Zenity, WitnessAI and Pillar Security; in part: Microsoft, Harmonic Security, Aim Security, CrowdStrike, Noma Security, Knostic, SentinelOne, Koi, Lasso Security and Palo Alto Networks. - **Credential access:** stops it: Bay, Zenity, Knostic and Pillar Security; in part: Microsoft, Harmonic Security, CrowdStrike, Noma Security, Island, SentinelOne, WitnessAI and Lasso Security. - **Data paste:** stops it: Microsoft, Harmonic Security, Aim Security, CrowdStrike, Island, SentinelOne, WitnessAI, Lasso Security, LayerX and Palo Alto Networks; in part: Bay, Noma Security, Zenity, Knostic and Koi. - **Package install:** stops it: CrowdStrike; in part: Bay, Island, Zenity, Knostic, Koi and LayerX. ## The ranked index 16 tools on seven criteria, weights in the header row. Each name opens the profile; the [full index](/ranking/) gives every score with its reason. | Rank | Tool | Total | Sees (25) | Stops (25) | Deployment (15) | Beyond (10) | Policy (10) | Transparency (10) | Proof (5) | | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | | 1 | [Microsoft](/platforms/microsoft/) | 75.0 | 8 | 8 | 6 | 8 | 9 | 7 | 4 | | 2 | [Bay](/platforms/bay/) | 71.5 | 8 | 8 | 9 | 7 | 6 | 1 | 8 | | 3 | [Harmonic Security](/platforms/harmonic-security/) | 71.5 | 8 | 7 | 7 | 7 | 7 | 6 | 7 | | 4 | [Aim Security](/platforms/aim-security/) | 69.0 | 8 | 7 | 5 | 7 | 6 | 7 | 8 | | 5 | [CrowdStrike](/platforms/crowdstrike/) | 66.0 | 8 | 7 | 6 | 6 | 7 | 3 | 7 | | 6 | [Noma Security](/platforms/noma-security/) | 66.0 | 8 | 8 | 6 | 5 | 5 | 4 | 6 | | 7 | [Island](/platforms/island/) | 66.0 | 7 | 7 | 6 | 8 | 7 | 3 | 8 | | 8 | [Zenity](/platforms/zenity/) | 65.0 | 6 | 7 | 5 | 7 | 7 | 7 | 8 | | 9 | [Knostic](/platforms/knostic/) | 64.0 | 7 | 7 | 7 | 3 | 6 | 7 | 5 | | 10 | [SentinelOne](/platforms/sentinelone/) | 62.0 | 7 | 7 | 5 | 7 | 6 | 4 | 5 | | 11 | [WitnessAI](/platforms/witnessai/) | 61.0 | 8 | 6 | 5 | 6 | 6 | 3 | 7 | | 12 | [Koi](/platforms/koi/) | 54.5 | 7 | 5 | 6 | 2 | 4 | 6 | 7 | | 13 | [Lasso Security](/platforms/lasso-security/) | 53.0 | 6 | 5 | 5 | 5 | 5 | 4 | 8 | | 14 | [Pillar Security](/platforms/pillar-security/) | 53.0 | 6 | 6 | 6 | 2 | 5 | 4 | 6 | | 15 | [LayerX](/platforms/layerx/) | 50.0 | 5 | 5 | 5 | 5 | 6 | 4 | 5 | | 16 | [Palo Alto Networks](/platforms/palo-alto-networks/) | 45.0 | 3 | 5 | 4 | 6 | 5 | 4 | 8 | 16 AI endpoint security tools by total (0 to 100), sorted on the published weights. | Tied at | Tools in order | Sees (heaviest) | What sets the order | | --- | --- | --- | --- | | 71.5 | Bay, Harmonic Security | 8, 8 | name order, as the scores are equal | | 66.0 | CrowdStrike, Noma Security, Island | 8, 8, 7 | name order, as the scores are equal | | 53.0 | Lasso Security, Pillar Security | 6, 6 | name order, as the scores are equal | Tied totals break on the higher score on the heaviest criterion, then by name. ## Who should pick what - [Coding agents](/best-for/coding-agents/): Bay 75.5; runner-up Microsoft 75.5 - [Regulated data](/best-for/regulated-data/): Microsoft 79.0; runner-up Harmonic Security 72.5 - [Lean team with EDR](/best-for/lean-team-edr/): Bay 76.5; runner-up Microsoft 73.5 - [Staff rollout](/best-for/staff-rollout/): Microsoft 75.5; runner-up Bay 73.5 - [Board in 30 days](/best-for/board-30-days/): Bay 77.5; runner-up Harmonic Security 73.5 ## What would change this ranking Microsoft is 3.5 points clear of Bay on the published weights. The smallest single change that would put Bay first is a Policy score of 10 instead of 6 (weight 10). Scores move only when a sourced fact changes ([how](/method/)). ## Where each is strongest Seesweight 258Microsoft, Bay, Harmonic Security, Aim Security, CrowdStrike, Noma Security, WitnessAIStopsweight 258Microsoft, Bay, Noma SecurityDeploymentweight 159BayBeyondweight 108Microsoft, IslandPolicyweight 109MicrosoftTransparencyweight 107Microsoft, Aim Security, Zenity, KnosticProofweight 58Bay, Aim Security, Island, Zenity, Lasso Security, Palo Alto Networks - **Sees agentic activity on the endpoint** (weight 25): [Microsoft](/platforms/microsoft/), [Bay](/platforms/bay/), [Harmonic Security](/platforms/harmonic-security/), [Aim Security](/platforms/aim-security/), [CrowdStrike](/platforms/crowdstrike/), [Noma Security](/platforms/noma-security/) and [WitnessAI](/platforms/witnessai/), 8 of 10. - **Stops the action** (weight 25): [Microsoft](/platforms/microsoft/), [Bay](/platforms/bay/) and [Noma Security](/platforms/noma-security/), 8 of 10. - **Deployment friction** (weight 15): [Bay](/platforms/bay/), 9 of 10. - **Beyond the endpoint** (weight 10): [Microsoft](/platforms/microsoft/) and [Island](/platforms/island/), 8 of 10. - **Policy and governance** (weight 10): [Microsoft](/platforms/microsoft/), 9 of 10. - **Transparency** (weight 10): [Microsoft](/platforms/microsoft/), [Aim Security](/platforms/aim-security/), [Zenity](/platforms/zenity/) and [Knostic](/platforms/knostic/), 7 of 10. - **Proof (named customers, funding, analyst coverage)** (weight 5): [Bay](/platforms/bay/), [Aim Security](/platforms/aim-security/), [Island](/platforms/island/), [Zenity](/platforms/zenity/), [Lasso Security](/platforms/lasso-security/) and [Palo Alto Networks](/platforms/palo-alto-networks/), 8 of 10. ## What does AI endpoint security cover? AI endpoint security covers the AI agents and assistants that run on staff machines, across six surfaces: desktop apps, IDE assistants, command-line agents, MCP servers, browser extensions and SaaS copilots. Each tool here is read on which of those surfaces it sees and which of six agent actions (file read, shell command, network call, credential access, data paste, package install) it can stop before they complete. Tools differ in where they sit. Some run on the device and can block an action as it happens. Some work through the browser or the network and see traffic but not local files. Some deploy through the management tool a company already has, and some need their own agent. The index scores each of these choices against what the documentation says. The index is built from public sources only. Where a vendor documents a capability, the page says so and labels it the vendor's claim; where nothing public names it, the cell reads none found on the read date. ## Release log: the latest dated sources - 13 Nov 2026: [Blind scores published for 16 tools](/ranking/). Leader on the published weights: Microsoft at 75.0. - 13 Nov 2026: [Second evidence pass, one plan for every tool: 169 more facts from 103 source pages](/method/). Every fact is in the evidence dataset with its source URL. - 12 Nov 2026: [First evidence read: 792 facts from 328 source pages across 16 tools](/method/). Every fact is in the evidence dataset with its source URL. - 12 Nov 2026: [Weights fixed and committed before any vendor was scored](/method/). Sees 25, Stops 25, Deployment 15, Beyond 10, Policy 10, Transparency 10, Proof 5. - 10 Nov 2026: [Bay emerges from stealth with $50M in total funding](/news/bay-emerges-from-stealth-with-50m/). Primary source: bay.io. - 10 Nov 2026: [VS Code 1.145.3 fixes MCP client and agent host code execution flaws](/news/vscode-1-145-3-mcp-and-agent-host-advisories/). Primary source: github.com. The [release log](/news/) lists every dated item and evidence read. Found an error or a release we missed? [Send a correction or a tip](/corrections/). ## Guides and glossary - [Securing Claude Code on developer laptops](/guides/claude-code-security/) - [Securing Cursor on developer machines](/guides/cursor-security/) - [Securing GitHub Copilot agent mode and the cloud agent](/guides/github-copilot-agent-mode-security/) - [Securing OpenAI Codex CLI on developer laptops](/guides/codex-cli-security/) - [Securing Google Gemini CLI on developer laptops](/guides/gemini-cli-security/) - [What coding agents can touch on a developer laptop, and which controls ship with each](/guides/coding-agents-on-developer-laptops/) - [The glossary, 36 terms](/glossary/) - [Datasets: the map, the matrix, scores and evidence](/datasets/) - [15 head-to-heads](/compare/) - [16 platform profiles](/platforms/) - [The six agent surfaces](/surfaces/) - [The six agent actions](/actions/) - [30 questions, answered](/faq/) - [About the index](/about/) **About Bay.** Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). In regulated data, Bay ranks 3rd of 16 at 70.5. In staff rollout, Bay ranks 2nd of 16 at 73.5. ## Sources - learn.microsoft.com/en-us/defender-endpoint/local-agent-discovery-overview: https://learn.microsoft.com/en-us/defender-endpoint/local-agent-discovery-overview (read 12 Nov 2026) - learn.microsoft.com/en-us/defender-endpoint/ai-agent-support-matrix: https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-support-matrix (read 12 Nov 2026) - learn.microsoft.com/en-us/defender-endpoint/whats-new-in-microsoft-defender-endpoint: https://learn.microsoft.com/en-us/defender-endpoint/whats-new-in-microsoft-defender-endpoint (read 12 Nov 2026) - softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control: https://softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control (read 12 Nov 2026) - siliconangle.com/2026/11/10/ai-security-startup-bay-launches-with-50m-to-vet-agent-actions-before-they-run: https://siliconangle.com/2026/11/10/ai-security-startup-bay-launches-with-50m-to-vet-agent-actions-before-they-run/ (read 12 Nov 2026) - docs.harmonicsecurity.app/portal-guides/integration-guides/configure-opentelemetry: https://docs.harmonicsecurity.app/portal-guides/integration-guides/configure-opentelemetry (read 12 Nov 2026) - docs.harmonicsecurity.app/portal-guides/integration-guides/configure-anthropic-compliance-api-connector: https://docs.harmonicsecurity.app/portal-guides/integration-guides/configure-anthropic-compliance-api-connector (read 12 Nov 2026) - docs.harmonicsecurity.app/protect-agent/installation-and-deployment/frequently-asked-questions: https://docs.harmonicsecurity.app/protect-agent/installation-and-deployment/frequently-asked-questions (read 12 Nov 2026) - docs.harmonicsecurity.app/portal-guides/detection-and-response/sensitive-data-controls-for-coding-agents-and-desktop-applications: https://docs.harmonicsecurity.app/portal-guides/detection-and-response/sensitive-data-controls-for-coding-agents-and-desktop-applications (read 12 Nov 2026) - AI Endpoint Security Ranked datasets (CSV and JSON): https://aiendpointsecurityranked.com/datasets/ (read 13 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/ --- # Microsoft leads 16 AI endpoint security tools at 75.0, 3.5 points ahead of Bay RANKED INDEX, read 13 Nov 2026 **The short answer.** Microsoft leads the index at 75.0 of 100. Bay follows at 71.5 and Harmonic Security at 71.5, and Palo Alto Networks closes the list at 45.0. Bay and Harmonic Security tie at 71.5 and both score 8 on Sees, the heaviest criterion here, so name order puts Bay 2nd. ## Which AI endpoint security tool ranks first? Microsoft ranks first of 16 AI endpoint security tools at 75.0 of 100, 3.5 points ahead of Bay at 71.5. Bay ranks 2nd of 16 at 71.5. The order comes from the published weights and the blind scores. ## How are AI endpoint security tools scored? Each tool gets seven scores from 0 to 10, one per criterion, and every score rests on sourced facts in the evidence file. The total runs 0 to 100 on weights fixed 12 Nov 2026 before any vendor was scored: Sees 25, Stops 25, Deployment 15, Beyond 10, Policy 10, Transparency 10, Proof 5. The [method](/method/) gives the rules and the score anchors. ## The full index | Rank | Tool | Total | Sees (25) | Stops (25) | Deployment (15) | Beyond (10) | Policy (10) | Transparency (10) | Proof (5) | | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | | 1 | [Microsoft](/platforms/microsoft/) | 75.0 | 8 | 8 | 6 | 8 | 9 | 7 | 4 | | 2 | [Bay](/platforms/bay/) | 71.5 | 8 | 8 | 9 | 7 | 6 | 1 | 8 | | 3 | [Harmonic Security](/platforms/harmonic-security/) | 71.5 | 8 | 7 | 7 | 7 | 7 | 6 | 7 | | 4 | [Aim Security](/platforms/aim-security/) | 69.0 | 8 | 7 | 5 | 7 | 6 | 7 | 8 | | 5 | [CrowdStrike](/platforms/crowdstrike/) | 66.0 | 8 | 7 | 6 | 6 | 7 | 3 | 7 | | 6 | [Noma Security](/platforms/noma-security/) | 66.0 | 8 | 8 | 6 | 5 | 5 | 4 | 6 | | 7 | [Island](/platforms/island/) | 66.0 | 7 | 7 | 6 | 8 | 7 | 3 | 8 | | 8 | [Zenity](/platforms/zenity/) | 65.0 | 6 | 7 | 5 | 7 | 7 | 7 | 8 | | 9 | [Knostic](/platforms/knostic/) | 64.0 | 7 | 7 | 7 | 3 | 6 | 7 | 5 | | 10 | [SentinelOne](/platforms/sentinelone/) | 62.0 | 7 | 7 | 5 | 7 | 6 | 4 | 5 | | 11 | [WitnessAI](/platforms/witnessai/) | 61.0 | 8 | 6 | 5 | 6 | 6 | 3 | 7 | | 12 | [Koi](/platforms/koi/) | 54.5 | 7 | 5 | 6 | 2 | 4 | 6 | 7 | | 13 | [Lasso Security](/platforms/lasso-security/) | 53.0 | 6 | 5 | 5 | 5 | 5 | 4 | 8 | | 14 | [Pillar Security](/platforms/pillar-security/) | 53.0 | 6 | 6 | 6 | 2 | 5 | 4 | 6 | | 15 | [LayerX](/platforms/layerx/) | 50.0 | 5 | 5 | 5 | 5 | 6 | 4 | 5 | | 16 | [Palo Alto Networks](/platforms/palo-alto-networks/) | 45.0 | 3 | 5 | 4 | 6 | 5 | 4 | 8 | All 16 tools on the published weights, total on a 0 to 100 scale. | Tied at | Tools in order | Sees (heaviest) | What sets the order | | --- | --- | --- | --- | | 71.5 | Bay, Harmonic Security | 8, 8 | name order, as the scores are equal | | 66.0 | CrowdStrike, Noma Security, Island | 8, 8, 7 | name order, as the scores are equal | | 53.0 | Lasso Security, Pillar Security | 6, 6 | name order, as the scores are equal | Tied totals break on the higher score on the heaviest criterion, then by name. Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). Bay is a young company that left stealth in 2026, publishes no price and has little independent testing behind its claims. ## Who should pick what - [Coding agents](/best-for/coding-agents/) (an engineering organization where every developer runs coding agents with shell access): pick Bay at 75.5, level with Microsoft and first on the tie-break. - [Regulated data](/best-for/regulated-data/) (a regulated company that must show which AI tools touch customer data): pick Microsoft at 79.0, with Harmonic Security next at 72.5. Bay ranks 3rd of 16 at 70.5. - [Lean team with EDR](/best-for/lean-team-edr/) (a lean security team with EDR in place and no view of AI agents): pick Bay at 76.5, with Microsoft next at 73.5. - [Staff rollout](/best-for/staff-rollout/) (a company rolling out an AI assistant and browser agents to all staff): pick Microsoft at 75.5, with Bay next at 73.5. Bay ranks 2nd of 16 at 73.5. - [Board in 30 days](/best-for/board-30-days/) (a CISO asked by the board for an inventory of AI agents within 30 days): pick Bay at 77.5, with Harmonic Security next at 73.5. ## What would change this ranking Microsoft is 3.5 points clear of Bay on the published weights. The smallest single change that would put Bay first is a Policy score of 10 instead of 6 (weight 10). Scores move only when a sourced fact changes ([how](/method/)). ## Every score, with the reason Each tool below lists its seven scores out of 10 with the reason in one line. A vendor's own claim scores at most 7 until a second source confirms it. ### 1. [Microsoft](/platforms/microsoft/), 75.0 | Criterion | Weight | Score | Why this score | | --- | --- | --- | --- | | Sees | 25 | 8 | Observed docs list named desktop, CLI and IDE agents with MCP configuration and VS Code extension discovery on Windows and macOS, and the support matrix now adds more agents. The vendor says this is not a complete inventory, and a skills inventory was none found on 13 Nov 2026. | | Stops | 25 | 8 | Observed docs show blocking of agent actions before they run for named CLI agents, including stopping a file's contents from being sent to an outside URL. The feature is a Windows-only preview, the vendor says intervention points vary by agent type, and package-install blocking was none found on 12 Nov 2026. | | Deployment | 15 | 6 | The feature uses the existing client, with Windows and macOS documented, but needs specific licences and active antivirus. No time-to-deploy figure was found on 12 Nov 2026. | | Beyond | 10 | 8 | Observed docs extend protections to SaaS copilots, ChatGPT and Claude enterprise, browser-detected AI apps and cloud-built agents. Some of this is an identity layer rather than the same endpoint control. | | Policy | 10 | 9 | Observed docs show SOC incidents, hunting tables, audit logs, SIEM and identity integration, and audit-to-block modes. Agent features are partly in preview. | | Transparency | 10 | 7 | Full public documentation with dated updates was observed, prices are published, and the docs state the licensing prerequisite. Published research beyond quoted incident remarks, and a readable third-party confirmation of price, were none found on 13 Nov 2026. | | Proof | 5 | 4 | The vendor quotes a few named customers, and third-party press confirms general availability. No product-specific analyst placement and no named customers in the docs were found on 12 Nov 2026. | Microsoft: total 75.0 of 100. ### 2. [Bay](/platforms/bay/), 71.5 | Criterion | Weight | Score | Why this score | | --- | --- | --- | --- | | Sees | 25 | 8 | Analyst and press coverage describe an inventory of agents, browser and IDE extensions, MCP servers, skills, plugins, models and packages. The analyst notes the breadth claims are vendor-reported. | | Stops | 25 | 8 | Analyst and press coverage report on-device allow, ask or deny before the action for files, network, credentials, MCP calls and shell. Prevention needs the on-device component, and latency and false-positive figures are vendor-reported with no test cited. | | Deployment | 15 | 9 | Third-party coverage reports agentless deployment through existing MDM or EDR on Windows, Mac and Linux in under five minutes. The deploy time is the vendor's own figure and has not been independently tested. | | Beyond | 10 | 7 | The vendor says one policy set covers devices, browsers, cloud workspaces and SaaS, and trade press restates coverage of devices, cloud and SaaS. The restatement names no browser or same-policy detail, so the detail rests on vendor claim. | | Policy | 10 | 6 | The vendor offers policies scoped by IdP group, OCSF events to SIEM and session records, and trade press says it integrates with security workflows. A named EDR integration and any independent confirmation of named integrations were none found on 13 Nov 2026. | | Transparency | 10 | 1 | The only public writing is a founder anecdote. Docs are sign-in only, and no price or published research was found on 12 Nov 2026. | | Proof | 5 | 8 | Third-party press confirms $50 million raised and names customers including Micron and Peloton, and an analyst firm profiles the vendor. Trade press notes one named customer is also an investor and that the vendor's Fortune 100 usage figure is unnamed and unverified. | Bay: total 71.5 of 100. ### 3. [Harmonic Security](/platforms/harmonic-security/), 71.5 | Criterion | Weight | Score | Why this score | | --- | --- | --- | --- | | Sees | 25 | 8 | The vendor names desktop apps, AI IDEs, command-line agents, local models and MCP servers, and observed docs show OTel ingestion and Compliance API visibility. No published inventory of skills or plugins was found on 12 Nov 2026, and coverage is set per customer. | | Stops | 25 | 7 | Observed docs and release notes show hook-based blocking of prompts and tool calls for many named coding agents, plus MCP gateway blocks, with limits written down, including that uploaded files are not covered. Separate blocked action types such as file reads, shell commands or credential access were none found on 12 Nov 2026, and an independent test was none found on 13 Nov 2026. | | Deployment | 15 | 7 | Observed docs list supported Windows, macOS and Linux versions and MDM installers with no proxy. A published time-to-deploy figure was none found on 12 Nov 2026, and an independent deployment source was none found on 13 Nov 2026. | | Beyond | 10 | 7 | The same platform covers browser tools, desktop apps, MCP agents and cloud-to-cloud agent connections, with cloud and SaaS paths observed in docs and release notes. The cloud piece is early access, and its docs say not every remote MCP server can be connected, naming six that do not work. | | Policy | 10 | 7 | Observed docs show identity-based access groups, Okta enrichment and webhook event types for SIEM; a full audit log is an add-on. A named EDR integration was none found on 12 Nov 2026, and an independent source on integrations was none found on 13 Nov 2026. | | Transparency | 10 | 6 | Public documentation and dated release notes were observed, and the vendor publishes usage research. No dollar price and no agent-vulnerability research were found on 12 Nov 2026. | | Proof | 5 | 7 | Third-party press confirms $50 million Series B funding and a regional distribution deal, the vendor names two customers, and it says Gartner lists it as an Example Vendor. The customer names and analyst mention are vendor-published only. | Harmonic Security: total 71.5 of 100. ### 4. [Aim Security](/platforms/aim-security/), 69.0 | Criterion | Weight | Score | Why this score | | --- | --- | --- | --- | | Sees | 25 | 8 | The vendor names local agents with their MCP servers and skills, plus browser and desktop AI, and observed product updates show inventory pages for agents, tools, MCP servers and skills, coding agent sessions and MCP discovery for a named agent. Discovery coverage of IDE extensions is not detailed in the observed records. | | Stops | 25 | 7 | Observed product updates show Local Agent policy rules with Block, Engage and Monitor actions and a detector that blocks destructive commands before they delete files or data, alongside network blocking of AI apps. Blocking is limited to supported local agents, and named blocking of credential access or network calls was none found on 13 Nov 2026. | | Deployment | 15 | 5 | A light client is pushed with existing tools, the vendor claims go-live in weeks, and observed product updates tie the AI traffic mode to Windows Client v6.4 and higher. A macOS or full operating-system list for the AI endpoint client was none found on 13 Nov 2026. | | Beyond | 10 | 7 | The vendor says the same engine and rules cover the browser, network AI traffic, managed agents and cloud workloads, and observed product updates show SaaS posture checks expanding to over 20 apps. The observed piece is posture checking rather than the same enforcement policy, so the cross-surface policy rests on vendor claim. | | Policy | 10 | 6 | The vendor describes allow, restrict and monitor policy with audit trails, and observed docs show policy configuration and a coding agent event type sent to third-party systems such as a SIEM. A documented EDR or identity-provider integration for AI events was none found on 13 Nov 2026. | | Transparency | 10 | 7 | Public platform docs and release notes were observed, and third-party sources confirm a CVE from its research. No published price and no AI endpoint-specific documentation page were found on 12 Nov 2026. | | Proof | 5 | 8 | Third-party press confirms funding, an acquisition and a named customer, and the vendor names further customers and a Gartner Leader placement. The analyst placement is platform-level, not specific to AI endpoints. | Aim Security: total 69.0 of 100. ### 5. [CrowdStrike](/platforms/crowdstrike/), 66.0 | Criterion | Weight | Score | Why this score | | --- | --- | --- | --- | | Sees | 25 | 8 | The vendor says it discovers agents, MCP servers, skills, desktop apps, browser extensions and IDEs, and trade press confirms the surfaces. Linux agent discovery is pre-beta. | | Stops | 25 | 7 | The vendor says hooks block harmful actions for named coding agents and that its sensor blocks malicious package installs before they run; trade press confirms prompt inspection. Third-party press reads the kill switch as gateway-only for now, and endpoint termination is still coming soon. | | Deployment | 15 | 6 | The vendor uses its existing lightweight sensor on Windows, macOS and Linux. No time-to-deploy figure was found on 12 Nov 2026. | | Beyond | 10 | 6 | The vendor enforces browser paste and upload policy, trade press confirms discovery of SaaS agents, and the vendor extends to a cloud agent guardrail service. Third-party press reports the cloud workspace integration is a macOS beta focused on visibility and detection without blocking, and SaaS agent coverage is discovery rather than enforcement. | | Policy | 10 | 7 | The vendor offers an agent graph for investigation, a gateway, managed response, audit-trail records of kill switch actions and its own SIEM, which trade press confirms ingests telemetry. An identity-provider integration was none found on 12 Nov 2026. | | Transparency | 10 | 3 | The vendor publishes research on agents and endpoint bundle prices, and trade press says the agent modules are sold through flexible licensing. Docs are login-gated, and a price for the AI product was none found on 12 Nov 2026. | | Proof | 5 | 7 | Third-party press confirms funding for the acquired startup, and the vendor names several customers and an IDC Leader placement. The analyst placement is for endpoint security, and analyst coverage of the AI product was none found on 13 Nov 2026. | CrowdStrike: total 66.0 of 100. ### 6. [Noma Security](/platforms/noma-security/), 66.0 | Criterion | Weight | Score | Why this score | | --- | --- | --- | --- | | Sees | 25 | 8 | The vendor names eight agents and inventories IDEs, MCP servers, skills, hooks and accounts from local files and EDR or MDM data, and an observed public repo hooks named agents. Browser coverage was none found on 12 Nov 2026. | | Stops | 25 | 8 | Observed plugin READMEs show block, mask and ask on shell commands, MCP calls, file reads and prompts. The hooks fail open, are in Beta, and some boundaries are marked coming soon. | | Deployment | 15 | 6 | Discovery uses existing EDR or MDM, and observed READMEs list macOS, Linux and Windows for the hooks plugin. The vendor says Windows is a one-by-one install today, the observed MDM path covers only macOS and Windows fleets, and a time-to-deploy figure was none found on 13 Nov 2026. | | Beyond | 10 | 5 | The vendor sells one control plane across endpoint, SaaS and cloud agents, but documents that its SaaS agent integration for one major platform is monitoring only, after the fact, with no inline blocking. Browser extension or browser-agent control was none found on 13 Nov 2026. | | Policy | 10 | 5 | The vendor gives each agent, MCP server and skill a registry status, limits actions by role, and ties policy to IdP users and groups; discovery uses the existing EDR or MDM. Audit trail detail and a named SIEM integration were none found on 13 Nov 2026. | | Transparency | 10 | 4 | The vendor publishes agent-vulnerability research, with several findings confirmed by third-party press. A published price and public documentation (docs redirect to a login page) were none found on 12 Nov 2026. | | Proof | 5 | 6 | Third-party press confirms a $100 million Series B and paying Fortune 500 customers, and the vendor cites Gartner placements. No named customers were found on 12 Nov 2026. | Noma Security: total 66.0 of 100. ### 7. [Island](/platforms/island/), 66.0 | Criterion | Weight | Score | Why this score | | --- | --- | --- | --- | | Sees | 25 | 7 | The vendor describes an endpoint inventory of agents, MCP servers, skills, packages, browser and IDE extensions and hooks, with named coding agents. Third-party coverage names the product lines for desktop apps and local agentic resources, but a second source confirming the inventory detail was none found on 13 Nov 2026. | | Stops | 25 | 7 | The vendor names blocking of Bash execution, file reads and outbound connections before an agent acts, plus gateway mitigation of injection. Third-party coverage confirms only human paste and download controls, and an independent test of agent-action blocking was none found on 13 Nov 2026. | | Deployment | 15 | 6 | The vendor pushes a Windows service and macOS daemon via MDM, and a customer reports a rollout of a couple of weeks. A Linux build, a vendor-published deploy time and an independent source on deployment were none found on 13 Nov 2026. | | Beyond | 10 | 8 | The vendor says the same policy reaches consumer browsers, agent browser sessions in the cloud, MCP calls and a SaaS agent platform, and third-party press reports policy enforced when endpoint or cloud agents call the browser, with data, credential and output controls. The third-party piece restates the vendor's launch rather than testing it. | | Policy | 10 | 7 | The vendor documents one policy framework for humans and agents, audit logs that tell them apart, gateway logs exportable to SIEM, and EDR and MDM integrations. All of it is vendor claim; an independent source on the integrations was none found on 13 Nov 2026. | | Transparency | 10 | 3 | The vendor publishes agent supply-chain and lookalike-domain research. A public pricing page and a public documentation site were none found on 12 Nov 2026. | | Proof | 5 | 8 | Third-party press confirms more than $1 billion raised and about 450 customers, and the vendor names customers such as Pfizer and TaskUs. Third-party press also confirms the Gartner mention rests on peer reviews, not an analyst ranking. | Island: total 66.0 of 100. ### 8. [Zenity](/platforms/zenity/), 65.0 | Criterion | Weight | Score | Why this score | | --- | --- | --- | --- | | Sees | 25 | 6 | The vendor names coding agents, inventories MCP connections and permission settings, and sees skills and plugins through a SaaS API. Browser extension or desktop app inventory, and an independent source for endpoint discovery, were none found on 13 Nov 2026. | | Stops | 25 | 7 | The vendor says hooks block tool calls before they run, naming sensitive file reads, network actions, secret exfiltration and malicious package installs. No independent test or public documentation confirming this was found on 12 Nov 2026. | | Deployment | 15 | 5 | The vendor says coding-agent coverage is agentless through hooks and OpenTelemetry. An operating-system list and a time-to-deploy figure were none found on 13 Nov 2026. | | Beyond | 10 | 7 | The vendor says one policy spans SaaS agents, cloud-built agents and coding agents, and adds inline control of a shared agent in Slack. Browser coverage and an independent confirmation of the shared policy were none found on 13 Nov 2026. | | Policy | 10 | 7 | The vendor lists Okta, EDR and SIEM integrations, SSO, deterministic policy and audit trails, and an alliance membership. All of it is vendor-stated, and an independent or public-docs confirmation was none found on 13 Nov 2026. | | Transparency | 10 | 7 | The vendor runs a documentation site and publishes agent-vulnerability research that third-party press covers, including a new notice on a cloud agent platform. A published price was none found on 12 Nov 2026. | | Proof | 5 | 8 | Third-party press confirms a $125 million round, the vendor names customers such as SoftBank on the record, and it cites Gartner reports including a Market Shaper placement. The customer names and analyst citations are vendor-published. | Zenity: total 65.0 of 100. ### 9. [Knostic](/platforms/knostic/), 64.0 | Criterion | Weight | Score | Why this score | | --- | --- | --- | --- | | Sees | 25 | 7 | The vendor names desktop, IDE and command-line agents and MCP servers, sees browser chatbots via network traffic, and open-source detection scripts were observed for one agent. Its own pages disagree on which agents are supported, and a second source for the full inventory was none found on 13 Nov 2026. | | Stops | 25 | 7 | The vendor says Enforce mode blocks secret file reads, destructive commands and risky MCP servers before they run; an observed open-source plugin blocks tool calls, but it is Alpha and for one agent. Shadow AI discovery is visibility-only, and a second-source test of the main product was none found on 13 Nov 2026. | | Deployment | 15 | 7 | The vendor publishes a light, MDM-pushed client for macOS and Windows (Linux beta) and claims deployment in minutes. All of it is vendor claim; an independent confirmation was none found on 13 Nov 2026. | | Beyond | 10 | 3 | Coverage beyond coding agents is a plugin for one non-coding agent plus visibility-only discovery of browser chatbots. No browser, cloud workspace or SaaS agent enforcement under the same policy was found on 12 Nov 2026. | | Policy | 10 | 6 | The vendor documents Off/Monitor/Enforce policies, audit logs, SIEM streaming and SSO, and an observed telemetry repo forwards to SIEM. No named EDR integration was found on 12 Nov 2026. | | Transparency | 10 | 7 | The vendor publishes list prices and a steady run of MCP research, and observed open-source repositories back the research. A readable public product documentation site was none found on 12 Nov 2026. | | Proof | 5 | 5 | Third-party press confirms an $11 million Series A and list awards, and the vendor says it was a Gartner Cool Vendor. No named customers were found on 12 Nov 2026. | Knostic: total 64.0 of 100. ### 10. [SentinelOne](/platforms/sentinelone/), 62.0 | Criterion | Weight | Score | Why this score | | --- | --- | --- | --- | | Sees | 25 | 7 | The vendor says it discovers AI in browsers, IDEs, terminal assistants and MCP servers, with skills and connectors, and names code assistants it tracks; third-party coverage confirms MCP discovery and a browser extension was observed. The full endpoint inventory rests on vendor claims. | | Stops | 25 | 7 | The vendor says it blocks credential exfiltration, tool poisoning and malicious MCP servers before actions run, and third-party coverage reports it can stop agents sending data out. The newest blocking feature is early access, and the usage-control product also offers coaching instead of blocking. | | Deployment | 15 | 5 | The endpoint needs a browser extension plus an endpoint agent, macOS and Windows are announced for one integration, and the vendor says the usage-control product deploys in minutes. A full OS list for the main endpoint agent was none found on 12 Nov 2026. | | Beyond | 10 | 7 | The vendor says coverage spans browsers, desktop apps, APIs and SaaS agent platforms, and that policy is enforced consistently across browser, desktop and embedded copilots. All of it is vendor claim. | | Policy | 10 | 6 | The vendor offers allow and block rules by user, agent and action, a full audit trail, an EDR execution graph and a customer-reported tie to identity systems. A named SIEM or identity-provider integration was none found on 12 Nov 2026. | | Transparency | 10 | 4 | The vendor publishes agent-risk research and free tools, and its public open-source repositories were observed. A published price and public documentation were none found on 12 Nov 2026. | | Proof | 5 | 5 | The vendor cites Gartner, Latio and an analyst-firm recognition, gives parent-company figures, and now names customers Carnival Corporation and 10x Banking in case studies. Product-specific funding and third-party confirmation of the customers were none found on 13 Nov 2026. | SentinelOne: total 62.0 of 100. ### 11. [WitnessAI](/platforms/witnessai/), 61.0 | Criterion | Weight | Score | Why this score | | --- | --- | --- | --- | | Sees | 25 | 8 | The vendor names desktop clients, IDEs, coding agents, plugins and MCP servers, and third-party coverage confirms discovery of desktop plugins, IDE extensions, local frameworks and MCP servers, plus an MCP catalog. Discovery works from network traffic, and an on-device inventory of skills or packages was none found on 12 Nov 2026. | | Stops | 25 | 6 | The vendor says it blocks unapproved MCP servers and tool calls inline at the network, and third-party coverage confirms prompt-injection blocking and credential tokenizing. Enforcement depends on traffic passing through it, and endpoint blocking of shell or file actions was none found on 12 Nov 2026. | | Deployment | 15 | 5 | The vendor deploys at the network level or through lightweight APIs with no endpoint client, and its CEO says one unnamed large customer was running in 3 days. An operating-system list was none found on 12 Nov 2026, and the deploy time is a single vendor anecdote. | | Beyond | 10 | 6 | The vendor says one policy covers humans, IDEs, chat apps and custom cloud agents, with orchestration-layer protection for custom agents. All of it is vendor claim, and browser-specific or named SaaS-agent coverage was none found on 13 Nov 2026. | | Policy | 10 | 6 | The vendor records user, agent, tool and rule for each block and ties identity to the IdP; third-party coverage confirms one policy and audit trail. A named EDR or SIEM integration was none found on 12 Nov 2026. | | Transparency | 10 | 3 | The vendor publishes security research on its blog and a self-published guardrail benchmark figure. Docs are password-gated and a published price was none found on 12 Nov 2026. | | Proof | 5 | 7 | Third-party press confirms more than $85 million raised, and the vendor names a customer, a reseller, an alliance membership and an analyst market map. An independent analyst rating or a customer named by a third party was none found on 13 Nov 2026. | WitnessAI: total 61.0 of 100. ### 12. [Koi](/platforms/koi/), 54.5 | Criterion | Weight | Score | Why this score | | --- | --- | --- | --- | | Sees | 25 | 7 | The vendor lists browser extensions, IDE plugins, MCP servers, packages and local models across endpoints with a real-time inventory, and third-party coverage describes software inventory and relays the vendor naming Claude Code, Codex CLI and OpenClaw among tools covered by the combined solution. A second source confirming the on-device agent inventory detail was none found on 13 Nov 2026. | | Stops | 25 | 5 | Observed docs say runtime protection intercepts policy-violating AI coding agent actions before they execute, trade press reports OS-level interception of prompts and responses, and application control blocks untrusted apps. The blocked agent action types are not named, which was none found on 13 Nov 2026. | | Deployment | 15 | 6 | Observed release notes list Windows, macOS and Linux, and observed docs and trade press describe a targeted sensor that runs beside existing third-party EDR under a separate license. A time-to-deploy figure was none found on 13 Nov 2026. | | Beyond | 10 | 2 | The vendor plans integration with a wider platform and has a software supply-chain gateway. No browser, cloud workspace or SaaS-agent coverage under one policy was found on 12 Nov 2026. | | Policy | 10 | 4 | The vendor ships out-of-the-box agent policies, observed docs show a session-centric audit trail of every coding agent action, and trade press reports policies mapped to MITRE ATLAS. SIEM export or identity-provider integration was none found on 13 Nov 2026. | | Transparency | 10 | 6 | The vendor's docs are public and third-party press covers its research, though one report is the subject of a defamation suit and was edited to remove the claimant. A published price was none found on 12 Nov 2026. | | Proof | 5 | 7 | Third-party press confirms $48 million raised and an acquisition by a large vendor, though reported deal values differ ($400 million against $231 million), and customer logos were observed. An independent analyst placement was none found on 13 Nov 2026. | Koi: total 54.5 of 100. ### 13. [Lasso Security](/platforms/lasso-security/), 53.0 | Criterion | Weight | Score | Why this score | | --- | --- | --- | --- | | Sees | 25 | 6 | The vendor inventories MCP connections for named desktop and coding agents, and an observed open-source scanner checks MCP servers. No browser extension inventory was found on 12 Nov 2026. | | Stops | 25 | 5 | The vendor says it blocks tool calls outside the allowed scope and risky MCP servers, and enforces some agent policies at the proxy or gateway layer; observed code shows gateway blocking and a warn-only hook. Named blocking of file reads, package installs or credential access was none found on 12 Nov 2026. | | Deployment | 15 | 5 | The vendor says it is agentless through the enterprise management platform and takes minutes to set up. An operating-system list and an independent deployment source were none found on 13 Nov 2026. | | Beyond | 10 | 5 | The vendor says one policy set covers SaaS chatbots, desktop agents and MCP, finds agents across SaaS, cloud and builder platforms, and reads a SaaS agent platform through an API. Browser extension coverage was none found on 12 Nov 2026, and an independent source was none found on 13 Nov 2026. | | Policy | 10 | 5 | The vendor offers audit trails, SIEM export of MCP logs and framework templates. A named EDR or identity-provider integration was none found on 12 Nov 2026, and an independent source was none found on 13 Nov 2026. | | Transparency | 10 | 4 | The vendor publishes research and open-source tools, confirmed by third-party and observed sources. Docs are login-gated and no published price was found on 12 Nov 2026. | | Proof | 5 | 8 | Third-party press confirms $30 million raised and names customers including BMW and a U.S. federal department, and reports a regional reseller; the vendor cites Gartner and Latio placements. The analyst placements are vendor-stated. | Lasso Security: total 53.0 of 100. ### 14. [Pillar Security](/platforms/pillar-security/), 53.0 | Criterion | Weight | Score | Why this score | | --- | --- | --- | --- | | Sees | 25 | 6 | The vendor names coding agents and inventories MCP servers, skills, plugins and hooks, with posture checks; it says it detects 18 coding agents but parses the configuration of only five. Desktop app or browser coverage and a second source were none found on 13 Nov 2026. | | Stops | 25 | 6 | The vendor says a pre-tool hook blocks credential reads followed by egress, public uploads, persistence writes, cloud metadata requests and unlisted MCP tools before they run. This is limited to three agents, the hooks fail open, and all of it is vendor claim. | | Deployment | 15 | 6 | The vendor deploys light scripts via MDM on macOS, Linux and Windows. A time-to-deploy figure and an independent deployment source were none found on 13 Nov 2026. | | Beyond | 10 | 2 | The wider platform covers homegrown AI and gateways. No browser, cloud workspace or SaaS-agent coverage under the endpoint policy was found on 12 Nov 2026. | | Policy | 10 | 5 | The vendor offers Off, Monitor and Block per control, a session timeline and session buffer for decisions, and a data API with Splunk and Jira integration. A named EDR or identity-provider integration was none found on 13 Nov 2026. | | Transparency | 10 | 4 | The vendor publishes agent-vulnerability research that third-party press credits, including a new MCP Inspector flaw. A published price and public documentation (access-code gate observed) were none found on 12 Nov 2026. | | Proof | 5 | 6 | Third-party press confirms the funding and an acquisition at about $150 million, and the vendor quotes named customers and cites Gartner recognition. The customer names and analyst mention are vendor-published. | Pillar Security: total 53.0 of 100. ### 15. [LayerX](/platforms/layerx/), 50.0 | Criterion | Weight | Score | Why this score | | --- | --- | --- | --- | | Sees | 25 | 5 | Third-party review confirms browser-session inventory of apps and extensions, the vendor names agentic browsers, and it says a native desktop agent covers desktop AI apps such as ChatGPT, Claude and Copilot desktop. Inventory of IDE assistants, command-line agents or MCP servers was none found on 12 Nov 2026. | | Stops | 25 | 5 | The vendor blocks prompts, pastes, uploads and risky extensions in the browser and says the same DLP applies on desktop, with the browser part confirmed by a third-party review. Blocking of agent shell commands, file reads, network calls or credential access was none found on 12 Nov 2026. | | Deployment | 15 | 5 | Deployment is a lightweight browser extension with no proxy, pushed by group policy or MDM per a third-party review, plus a desktop agent installed through standard tooling. An operating-system list and a time-to-deploy figure were none found on 13 Nov 2026. | | Beyond | 10 | 5 | The vendor covers browsers, SaaS apps and agentic browsers under its browser policy. Cloud workspace or SaaS-hosted agent coverage, and an independent confirmation, were none found on 13 Nov 2026. | | Policy | 10 | 6 | A central console, IdP and SIEM integrations, a cloud security console feed and per-identity attribution are listed, and a third-party review describes the policy model. An EDR integration was none found on 12 Nov 2026. | | Transparency | 10 | 4 | The vendor publishes agent and browser research that third-party press reports, and it sells through a cloud marketplace. A published price figure and public documentation (docs redirect to sign-in) were none found on 12 Nov 2026. | | Proof | 5 | 5 | Third-party press confirms seed and Series A funding, and the vendor reports an acquisition. No named customers or analyst ranking were found on 12 Nov 2026. | LayerX: total 50.0 of 100. ### 16. [Palo Alto Networks](/platforms/palo-alto-networks/), 45.0 | Criterion | Weight | Score | Why this score | | --- | --- | --- | --- | | Sees | 25 | 3 | The vendor documents an inventory of browser extensions and claims visibility into all AI apps and agentic workflows; observed docs show an AI agent discovery feature with first-phase limits on cloud data. Named desktop agents, IDE assistants, command-line agents or an endpoint MCP inventory were none found on 13 Nov 2026. | | Stops | 25 | 5 | The vendor blocks pastes, uploads and transfers to unsanctioned apps and claims coding-agent enforcement that helps stop destructive commands; trade press confirms network policy enforcement on DPUs is shipping, while a CPU gateway is forward-looking. A public page naming which agent actions are blocked before completion was none found on 12 Nov 2026. | | Deployment | 15 | 4 | The vendor ships a browser, extension or mobile app deployable from an email link without admin rights, plus network and API intercept modes. An operating-system list, endpoint agent footprint or time to deploy was none found on 13 Nov 2026. | | Beyond | 10 | 6 | The vendor says one browser policy controls SaaS, GenAI and private apps, and claims coverage across cloud, SaaS and endpoints, with SaaS agent prompts monitored rather than blocked. A partner integration for agent traffic has no release date, and an independent confirmation of one shared policy was none found on 13 Nov 2026. | | Policy | 10 | 5 | The vendor describes dynamic risk-based policies and audit trails across web actions, and trade press reports an integration that enforces network and API access rules and sends telemetry to the vendor's own SIEM platform. A named EDR or identity-provider integration was none found on 13 Nov 2026. | | Transparency | 10 | 4 | Public technical documentation, including a dated browser product docs site, and release notes were observed. A published price and published research were none found on 12 Nov 2026. | | Proof | 5 | 8 | Third-party press confirms an acquisition of more than $500 million, the vendor names customer voices at three companies, and it cites a Gartner placement. The analyst placement is vendor-stated. | Palo Alto Networks: total 45.0 of 100. ## Sources - Microsoft: learn.microsoft.com/en-us/defender-endpoint/local-agent-discovery-overview: https://learn.microsoft.com/en-us/defender-endpoint/local-agent-discovery-overview (read 12 Nov 2026) - Microsoft: learn.microsoft.com/en-us/defender-endpoint/ai-agent-support-matrix: https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-support-matrix (read 12 Nov 2026) - Bay: www.bay.io/platform/ai-endpoint-security: https://www.bay.io/platform/ai-endpoint-security (read 12 Nov 2026) - Bay: www.bay.io/platform: https://www.bay.io/platform (read 12 Nov 2026) - Harmonic Security: www.harmonic.security/solutions/endpoint-ai-security: https://www.harmonic.security/solutions/endpoint-ai-security (read 12 Nov 2026) - Aim Security: www.catonetworks.com/platform/cato-ai-endpoint: https://www.catonetworks.com/platform/cato-ai-endpoint/ (read 12 Nov 2026) - CrowdStrike: www.crowdstrike.com/en-us/platform/falcon-guardian-aidr: https://www.crowdstrike.com/en-us/platform/falcon-guardian-aidr/ (read 12 Nov 2026) - CrowdStrike: www.crowdstrike.com/en-us/blog/falcon-guardian-defines-next-generation-of-ai-security: https://www.crowdstrike.com/en-us/blog/falcon-guardian-defines-next-generation-of-ai-security/ (read 12 Nov 2026) - Noma Security: noma.security/platform/endpoint-agents: https://noma.security/platform/endpoint-agents (read 12 Nov 2026) - Noma Security: noma.security/blog/noma-brings-agent-security-to-every-employee-endpoint: https://noma.security/blog/noma-brings-agent-security-to-every-employee-endpoint (read 12 Nov 2026) - Island: www.island.io/product/ai: https://www.island.io/product/ai (read 12 Nov 2026) - Island: www.island.io/blog/expanding-island-to-govern-your-ai-workforce: https://www.island.io/blog/expanding-island-to-govern-your-ai-workforce (read 12 Nov 2026) - Zenity: zenity.io/use-cases/agent-type/coding-personal-agents: https://zenity.io/use-cases/agent-type/coding-personal-agents (read 12 Nov 2026) - Zenity: zenity.io/blog/coding-agent-attack-surface: https://zenity.io/blog/coding-agent-attack-surface (read 12 Nov 2026) - Knostic: www.knostic.ai/faq: https://www.knostic.ai/faq (read 12 Nov 2026) - Knostic: www.knostic.ai/kirin: https://www.knostic.ai/kirin (read 12 Nov 2026) - SentinelOne: www.sentinelone.com/blog/a-new-chapter-for-ai-and-cybersecurity-sentinelone-acquires-prompt-security: https://www.sentinelone.com/blog/a-new-chapter-for-ai-and-cybersecurity-sentinelone-acquires-prompt-security/ (read 12 Nov 2026) - SentinelOne: www.sentinelone.com/blog/prompt-security-for-agentic-ai: https://www.sentinelone.com/blog/prompt-security-for-agentic-ai/ (read 12 Nov 2026) - WitnessAI: witness.ai: https://witness.ai/ (read 12 Nov 2026) - WitnessAI: witness.ai/for-agents: https://witness.ai/for-agents/ (read 12 Nov 2026) - Koi: www.paloaltonetworks.com/resources/datasheets/cortex-agentic-endpoint-security-at-a-glance: https://www.paloaltonetworks.com/resources/datasheets/cortex-agentic-endpoint-security-at-a-glance (read 12 Nov 2026) - Koi: www.paloaltonetworks.com/resources/techbriefs/koi-agentic-endpoint-security: https://www.paloaltonetworks.com/resources/techbriefs/koi-agentic-endpoint-security (read 12 Nov 2026) - Lasso Security: www.lasso.security/platform/ai-usage-control: https://www.lasso.security/platform/ai-usage-control (read 12 Nov 2026) - Lasso Security: www.lasso.security/use-cases/mcp-security: https://www.lasso.security/use-cases/mcp-security (read 12 Nov 2026) - Pillar Security: www.pillar.security/platform/agentic-endpoint: https://www.pillar.security/platform/agentic-endpoint (read 12 Nov 2026) - LayerX: www.akamai.com/products/akamai-workforce-protector: https://www.akamai.com/products/akamai-workforce-protector (read 12 Nov 2026) - Palo Alto Networks: www.paloaltonetworks.com/prisma/prisma-ai-runtime-security: https://www.paloaltonetworks.com/prisma/prisma-ai-runtime-security (read 12 Nov 2026) - Palo Alto Networks: www.paloaltonetworks.com/sase/prisma-browser: https://www.paloaltonetworks.com/sase/prisma-browser (read 12 Nov 2026) - Scores dataset, CSV: https://aiendpointsecurityranked.com/datasets/aes-scores.csv (read 13 Nov 2026) - Evidence dataset, CSV: https://aiendpointsecurityranked.com/datasets/aes-evidence.csv (read 13 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/ranking/ --- # Weights fixed first, evidence second, scores last, each step dated METHOD, read 13 Nov 2026 **The short answer.** Seven criteria and their weights were committed on 12 Nov 2026, before any vendor was scored. A scorer given only the rubric and the evidence files then scored every tool, 0 to 10, and the total is the weighted sum on a 0 to 100 scale. Every step is dated. The index answers one question: which tool shows a security team the AI agents on its machines and stops the ones that go wrong. The seven criteria below are what a buyer asking that question weighs, written down before any evidence was gathered, so a reader can agree or disagree with the weights before seeing a score. Each tool is scored 0 to 10 on each criterion from its evidence file, a list of facts with a quote, a source URL and the date it was read. The weighted sum of the scores, divided by 10, is the total on a 0 to 100 scale. Ties break on the higher score on the heaviest criterion, then by name. ## The criteria and their weights | Criterion | What is scored, with evidence | Weight | | --- | --- | --- | | Sees agentic activity on the endpoint | An inventory of AI agents, skills, plugins, MCP servers and extensions across desktop apps, IDEs, command-line agents and browsers, as documented | 25 | | Stops the action | Blocking a harmful agent action before it completes, with the action types named, against detection or alerting after the fact | 25 | | Deployment friction | Agentless or light deployment, operating systems covered, time to deploy as published | 15 | | Beyond the endpoint | Coverage of browsers, cloud workspaces and SaaS-hosted agents in the same policy | 10 | | Policy and governance | The policy model, the audit trail, and integrations with EDR, SIEM and the identity provider | 10 | | Transparency | Public documentation, a published price, published research | 10 | | Proof | Named customers, funding, independent analyst coverage | 5 | Weights fixed 12 Nov 2026 and committed before the first score. The committed file is published as weights.json. ## What each score means - **Score 0:** none found on the read date - **Score 1-3:** mentioned, partial, or limited to one surface or one action type - **Score 4-6:** documented and working, with gaps the documentation shows - **Score 7:** the most a vendor claim or vendor documentation alone can earn: detailed, specific, covering the whole criterion - **Score 8-10:** detailed and confirmed by a second source (third-party write-up, or observed directly in public documentation, code or a release record) ## The rules every score follows - Every vendor is scored 0 to 10 on every criterion, whole numbers, from the evidence file only, by a scorer given no brief about any vendor. - A vendor claim scores at most 7 until a second source confirms it (kind third-party or observed). - None found scores 0 and is written "none found on the read date", and is never written as a missing feature. - Breadth facts (module counts, integration counts, headcount, funding) carry at most 10 percent of any ranking; they inform Proof only, which weighs 5. - The total is the weighted sum of the scores divided by 10 (0 to 100 scale, published to one decimal). - Ties break on the higher score on the heaviest criterion, then by name. - Once the first score exists, no weight and no criterion changes. A score changes only when a sourced fact changes, logged with the fact. How the tie-break reads the heaviest criterion: where two criteria share the top weight, as Sees and Stops do on the site weights at 25 each, the first listed (Sees) counts as the heaviest: ties on the site weights break on Sees, then by name, which is how the published ranking script applied it, so stating it changes no order. Each situation has a single heaviest criterion. ## The five situations and their own weights Each situation page ranks the same scores under weights a buyer in that situation would use. They were committed together with the site weights, before any vendor was scored. | Situation | Sees | Stops | Deployment | Beyond | Policy | Transparency | Proof | | --- | --- | --- | --- | --- | --- | --- | --- | | [Coding agents](/best-for/coding-agents/) | 25 | 35 | 15 | 5 | 10 | 5 | 5 | | [Regulated data](/best-for/regulated-data/) | 30 | 15 | 5 | 15 | 25 | 5 | 5 | | [Lean team with EDR](/best-for/lean-team-edr/) | 30 | 20 | 25 | 5 | 10 | 5 | 5 | | [Staff rollout](/best-for/staff-rollout/) | 20 | 20 | 15 | 25 | 10 | 5 | 5 | | [Board in 30 days](/best-for/board-30-days/) | 40 | 5 | 30 | 10 | 5 | 5 | 5 | Situation weights, each set summing to 100. - **Coding agents:** A coding agent with a shell can delete, exfiltrate or install in seconds, so stopping the action before it completes matters most, followed by seeing which agents, MCP servers and extensions each developer runs. - **Regulated data:** An auditor asks for the inventory and the evidence trail, so seeing every AI tool and keeping a policy record with an audit log outweigh how fast the tool installs. - **Lean team with EDR:** A small team needs the missing view quickly without another heavy agent to run, so visibility and low deployment friction carry the most weight. - **Staff rollout:** Staff-wide assistants and browser agents live in the browser and in SaaS as much as on the laptop, so coverage beyond the endpoint under one policy weighs most. - **Board in 30 days:** A 30-day inventory deadline rewards the widest view of agents on the most machines, deployed fastest; blocking can follow later. ## Evidence pass of 12 Nov 2026 and the blind re-score On 12 Nov 2026 a second evidence pass added facts for every vendor, and all 112 cells (16 tools, 7 criteria) were then re-scored blind from the updated evidence files under the same weights. 70 scores changed; each change is listed with the facts it now cites in the [score change log](/datasets/score-changes.csv) (JSON: [score-changes.json](/datasets/score-changes.json)). Re-scope of Palo Alto Networks, 12 Nov 2026: 13 facts about Cortex Agentic Endpoint Security, which is the Koi product, were moved to the Koi profile because Koi has its own profile. Palo Alto Networks was then re-scored blind against the other 15 vendors' reviewed scores, and its total moved from 58.0 to 44.0. The three rows for that re-scope are at the end of the [score change log](/datasets/score-changes.csv). ## Evidence pass of 13 Nov 2026 On 13 Nov 2026 a further evidence pass ran one search plan for all 16 tools alike: the same news and analyst queries, the same source types and the same budget of 45 fetches per tool, aimed at second sources for scores at the 7-of-10 cap and for lower scores resting on vendor claims alone, anything published since the first read, and evidence against each current score. It added 134 facts from 78 new source pages, plus 35 records of a second source looked for and none found. The 87 cells those facts touched were re-scored blind and checked by a second blind reviewer under the same rubric and weights; 20 scores moved, 15 up and 5 down, and every other cell kept its score. | Tool | Criterion | Before | After | Moved by | | --- | --- | --- | --- | --- | | Aim Security | Deployment friction | 4 | 5 | aim-security-69 | | Aim Security | Policy and governance | 5 | 6 | aim-security-71 | | Aim Security | Sees agentic activity on the endpoint | 7 | 8 | aim-security-67; aim-security-66 | | Aim Security | Stops the action | 5 | 7 | aim-security-64; aim-security-65 | | Bay | Proof | 9 | 8 | bay-57; bay-56 | | CrowdStrike | Beyond the endpoint | 7 | 6 | crowdstrike-52 | | Harmonic Security | Beyond the endpoint | 8 | 7 | harmonic-security-53 | | Island | Beyond the endpoint | 7 | 8 | island-51 | | Koi (Cortex Agentic Endpoint Security) | Deployment friction | 2 | 6 | koi-48; koi-49; koi-55 | | Koi (Cortex Agentic Endpoint Security) | Policy and governance | 2 | 4 | koi-51; koi-56 | | Koi (Cortex Agentic Endpoint Security) | Stops the action | 3 | 5 | koi-50; koi-54 | | LayerX (now Akamai Workforce Protector) | Sees agentic activity on the endpoint | 4 | 5 | layerx-57 | | Noma Security | Beyond the endpoint | 6 | 5 | noma-security-55 | | Noma Security | Deployment friction | 7 | 6 | noma-security-52; noma-security-56 | | Palo Alto Networks | Policy and governance | 4 | 5 | palo-alto-networks-57 | | Pillar Security | Policy and governance | 4 | 5 | pillar-security-47 | | SentinelOne | Beyond the endpoint | 6 | 7 | sentinelone-57 | | SentinelOne | Deployment friction | 4 | 5 | sentinelone-56 | | SentinelOne | Proof | 4 | 5 | sentinelone-59; sentinelone-60 | | WitnessAI | Deployment friction | 4 | 5 | witnessai-58 | Every score the 13 Nov 2026 pass moved. Fact ids are in the evidence dataset. The Stops score and the Action Control Matrix measure different things: the matrix counts which of six action types a vendor documents, while the Stops score rates how specific and confirmed the blocking evidence is (action types named, blocking before completion, a second source), so a tool can document fewer action types and still score higher on Stops. ## How the scoring was kept blind Scoring was done from the evidence files by a scorer given the rubric and the files, with no brief about any vendor. The evidence for every vendor was gathered to the same template: the same six surfaces, the same six actions, the same questions on price, deployment and documentation. The weights file was committed before the first score file existed and has not changed since. The build checks this from the git history, checks that every ranked view and head-to-head on the site matches the scores file, and fails if any of them differ. The weights file carries the fixed date 12 Nov 2026. Scores are dated 13 Nov 2026. [Download the weights file](/datasets/weights.json), the same file the build reads. ## What the scores leave out The index reads public sources: vendor documentation, release notes, press and analyst coverage, and public code. It does not test products in a lab, so a documented capability is a statement about what the vendor publishes and a missing one is a statement about what could be found. Breadth facts such as headcount, funding and the number of integrations inform the Proof criterion only, which carries 5 of 100, and never more than 10 percent of any ranking. Short answers to 30 common questions are on the [FAQ page](/faq/). ## Changelog - 12 Nov 2026: the seven criteria, their weights and the five situation weightings fixed and committed before any evidence was gathered or any vendor scored. - 12 Nov 2026: blind scores for 16 tools, then an evidence pass and a blind re-score of every cell (70 scores changed, each in the [score change log](/datasets/score-changes.csv)). - 12 Nov 2026: Palo Alto Networks re-scoped, with the Cortex Agentic Endpoint Security facts moved to the Koi profile, and re-scored blind. - 13 Nov 2026: a further evidence pass on one plan for all 16 tools; 20 scores moved. - 16 Nov 2026: a new design and the answer layer (question headings, who should pick what, what would change this ranking, the [release log](/news/) and the action finder). No weight, score, ranking, vendor or page address changed. ## Sources - Committed weights file, fixed 12 Nov 2026 (JSON): https://aiendpointsecurityranked.com/datasets/weights.json (read 13 Nov 2026) - Evidence dataset, CSV: https://aiendpointsecurityranked.com/datasets/aes-evidence.csv (read 13 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/method/ --- # Microsoft: 1st of 16 at 75.0, strongest on Policy at 9 of 10, the highest in the index PLATFORM PROFILE, read 13 Nov 2026 **The short answer.** Microsoft ranks 1st of 16 AI endpoint security tools at 75.0 of 100. Microsoft stands out on Policy at 9 of 10, the highest in the index, and its lowest score is Proof (named customers, funding, analyst coverage) at 4 of 10. Scores run 0 to 100 on weights fixed before any vendor was scored. ## Profile - **Owner:** Microsoft Corporation - **Product:** Microsoft Defender for Endpoint AI agent discovery and runtime protection, with Agent 365, Purview DSPM and Endpoint DLP, and Entra Agent ID - **Status:** Active under this name, checked 13 Nov 2026 - **Category:** Platform vendor: endpoint security, data loss prevention and identity controls for AI agents - **Deployment:** Existing Defender for Endpoint client with Intune, PowerShell or Defender portal policy; Purview browser extension for AI sites - **Operating systems:** Windows, macOS (discovery and Endpoint DLP) - **Time to deploy:** none found on 12 Nov 2026 - **Price:** Microsoft 365 E5 $60.00 and E7 $99.00 per user per month, Agent 365 $15.00 per user per month, paid yearly; no separate price for Defender AI agent runtime protection found on 12 Nov 2026 (the vendor's own claim) - **Website:** [learn.microsoft.com](https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-runtime-protection-overview) Scores out of 10 (weight)Sees 258Stops 258Deployment 156Beyond 108Policy 109Transparency 107Proof 54 ## Strength and weakness **Strength:** Defender discovers dozens of local agents and MCP servers on Windows and macOS (GA) and can block prompt-injection-driven actions in Claude Code, Codex CLI and GitHub Copilot CLI on Windows, with Endpoint DLP blocking pastes and uploads to AI sites. **Weakness:** Runtime blocking is Windows-only preview with a documented Claude Code Block-mode limitation, and none found on 12 Nov 2026 for package-install blocking or a published time to deploy. ## Scores: 75.0 of 100, 1st of 16 | Criterion | Weight | Score | Why this score | | --- | --- | --- | --- | | Sees agentic activity on the endpoint | 25 | 8 | Observed docs list named desktop, CLI and IDE agents with MCP configuration and VS Code extension discovery on Windows and macOS, and the support matrix now adds more agents. The vendor says this is not a complete inventory, and a skills inventory was none found on 13 Nov 2026. | | Stops the action | 25 | 8 | Observed docs show blocking of agent actions before they run for named CLI agents, including stopping a file's contents from being sent to an outside URL. The feature is a Windows-only preview, the vendor says intervention points vary by agent type, and package-install blocking was none found on 12 Nov 2026. | | Deployment friction | 15 | 6 | The feature uses the existing client, with Windows and macOS documented, but needs specific licences and active antivirus. No time-to-deploy figure was found on 12 Nov 2026. | | Beyond the endpoint | 10 | 8 | Observed docs extend protections to SaaS copilots, ChatGPT and Claude enterprise, browser-detected AI apps and cloud-built agents. Some of this is an identity layer rather than the same endpoint control. | | Policy and governance | 10 | 9 | Observed docs show SOC incidents, hunting tables, audit logs, SIEM and identity integration, and audit-to-block modes. Agent features are partly in preview. | | Transparency | 10 | 7 | Full public documentation with dated updates was observed, prices are published, and the docs state the licensing prerequisite. Published research beyond quoted incident remarks, and a readable third-party confirmation of price, were none found on 13 Nov 2026. | | Proof | 5 | 4 | The vendor quotes a few named customers, and third-party press confirms general availability. No product-specific analyst placement and no named customers in the docs were found on 12 Nov 2026. | Microsoft: weighted total 75.0 of 100 on the published weights. ## What does Microsoft see on an endpoint? Microsoft documents 5 of 6 agent surfaces in full (desktop apps, IDE assistants, command-line agents, MCP servers and SaaS copilots) and 1 in part (browser extensions). Each cell with its note and source: [Microsoft on the Agent Surface Map](/agent-surface-map/microsoft/). ## Which agent actions can Microsoft stop? Microsoft documents stopping 2 of 6 agent actions before they complete (shell command and data paste) and 3 in part (file read, network call and credential access). Nothing was found for package install. Each cell with its note and source: [Microsoft on the Action Control Matrix](/action-control-matrix/microsoft/). ## How much does Microsoft cost? Microsoft publishes a price: microsoft 365 E5 $60.00 and E7 $99.00 per user per month, Agent 365 $15.00 per user per month, paid yearly; no separate price for Defender AI agent runtime protection found on 12 Nov 2026. (the vendor's own claim) ## Evidence behind the scores - Microsoft docs say Defender discovers local AI agents and their configured MCP servers on onboarded devices and shows them in the Defender portal. (observed in public documentation or code; [learn.microsoft.com](https://learn.microsoft.com/en-us/defender-endpoint/local-agent-discovery-overview), read 12 Nov 2026) - The Defender support matrix lists Claude Code, Claude Desktop, Codex CLI, Codex Desktop and ChatGPT Desktop as discovered with MCP configuration discovery on Windows. (observed in public documentation or code; [learn.microsoft.com](https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-support-matrix), read 12 Nov 2026) - The Defender support matrix lists IDE and CLI assistants including Cursor, GitHub Copilot CLI, Gemini CLI and Windsurf as discovered with MCP discovery on Windows. (observed in public documentation or code; [learn.microsoft.com](https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-support-matrix), read 12 Nov 2026) - Microsoft release notes say local AI agent discovery on Windows and macOS reached general availability in October 2026, with WSL discovery still in preview. (observed in public documentation or code; [learn.microsoft.com](https://learn.microsoft.com/en-us/defender-endpoint/whats-new-in-microsoft-defender-endpoint), read 12 Nov 2026) - Defender collects installed VS Code extension data for agents like GitHub Copilot, Claude Code and Cline extensions on Windows and macOS, and the extension does not need to run. (observed in public documentation or code; [learn.microsoft.com](https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-support-matrix), read 12 Nov 2026) - VentureBeat reports Agent 365, powered by Defender and Intune, detects OpenClaw agents on managed Windows devices and plans broader local agent discovery, including GitHub Copilot CLI and Claude Code. (third-party source; [venturebeat.com](https://venturebeat.com/technology/microsoft-takes-agent-365-out-of-preview-as-shadow-ai-becomes-an-enterprise-threat), read 12 Nov 2026) - Microsoft states discovery is not a complete software inventory and covers only agents Defender observes. (observed in public documentation or code; [learn.microsoft.com](https://learn.microsoft.com/en-us/defender-endpoint/local-agent-discovery-overview), read 12 Nov 2026) - Microsoft's AI agent support matrix, updated 12 Nov 2026, now lists additional agents for discovery (Factory Droid CLI and Factory App, Kimi Work, Replit Agent), after the 3 Nov 2026 version held in this file. (observed in public documentation or code; [learn.microsoft.com](https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-support-matrix), read 13 Nov 2026) - Microsoft says Defender runtime protection detects malicious prompts and harmful agent actions, audits them, and blocks actions before they run when the inspection method allows it. (observed in public documentation or code; [learn.microsoft.com](https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-runtime-protection-overview), read 12 Nov 2026) - Microsoft says agent-native inspection uses vendor event interfaces for Claude Code, Codex CLI and GitHub Copilot CLI, applying audit or block decisions at points in the agent workflow. (observed in public documentation or code; [learn.microsoft.com](https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-runtime-protection-overview), read 12 Nov 2026) - Microsoft documents an example where Defender detects a prompt injection in a tool response and blocks an attempt to send local environment file contents to an outside URL before data leaves the device. (observed in public documentation or code; [learn.microsoft.com](https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-runtime-protection-overview), read 12 Nov 2026) - Microsoft's runtime protection support matrix is labeled preview and lists Windows only, with agent-native inspection for Claude Code, GitHub Copilot CLI and app, Claude Desktop Code, and VS Code extensions. (observed in public documentation or code; [learn.microsoft.com](https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-support-matrix), read 12 Nov 2026) - Microsoft lists a preview known issue: Claude Code may keep processing a tool response even in Block mode, and post-tool-response inspection cannot undo actions already completed. (observed in public documentation or code; [learn.microsoft.com](https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-runtime-protection-overview), read 12 Nov 2026) - Microsoft's own security blog says Defender real-time protection coverage and intervention points differ by agent type, a limit on how uniformly agent actions can be blocked. (the vendor's own claim; [microsoft.com](https://www.microsoft.com/en-us/security/blog/2026/11/12/trained-to-betray-how-post-training-supply-chain-attacks-hijack-coding-agents/), read 13 Nov 2026) ## Checks that found nothing - Microsoft documentation for blocking malicious package installs by AI agents in the Defender AI agent runtime protection or support matrix pages: none found on 12 Nov 2026 (checked: [learn.microsoft.com](https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-support-matrix)) - A published time-to-deploy figure for Defender AI agent discovery or runtime protection; docs recommend a one-to-two-week audit-mode review before enforcement: none found on 12 Nov 2026 (checked: [learn.microsoft.com](https://learn.microsoft.com/en-us/defender-endpoint/configure-ai-agent-runtime-protection)) - A readable third-party page confirming the Agent 365 or Defender price; Google News listed trade headlines (Register, CRN, TechTarget, UC Today) but their pages could not be opened: none found on 13 Nov 2026 (checked: [news.google.com](https://news.google.com/rss/search?q=%22Agent+365%22+price+OR+pricing+per+user&hl=en-US&gl=US&ceid=US:en)) - Named customers or analyst reports specific to Defender AI agent discovery and runtime protection on the documentation pages read: none found on 12 Nov 2026 (checked: [learn.microsoft.com](https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-runtime-protection-overview)) - A Gartner, Forrester or IDC analyst placement specific to Defender AI agent discovery or Agent 365; Microsoft's analyst-reports page and guessed Gartner Magic Quadrant blog URLs returned 404 Page cannot be found, so no general endpoint-platform placement is cited here either: none found on 12 Nov 2026 (checked: [microsoft.com](https://www.microsoft.com/en-us/security/business/analyst-reports)) ## Position in each situation | Situation | Position | Total | Leader | | --- | --- | --- | --- | | [Coding agents](/best-for/coding-agents/) | 2nd of 16 (tied) | 75.5 | Bay 75.5 | | [Regulated data](/best-for/regulated-data/) | 1st of 16 | 79.0 | Microsoft 79.0 | | [Lean team with EDR](/best-for/lean-team-edr/) | 2nd of 16 | 73.5 | Bay 76.5 | | [Staff rollout](/best-for/staff-rollout/) | 1st of 16 | 75.5 | Microsoft 75.5 | | [Board in 30 days](/best-for/board-30-days/) | 3rd of 16 | 72.0 | Bay 77.5 | Microsoft under each situation's own weights. Tied marks an equal total; the order follows the tie-break, the higher score on the heaviest criterion, then name. ## Head-to-heads - [Bay vs Microsoft](/compare/bay-vs-microsoft/): Microsoft 75.0 to Bay 71.5 on the published weights - [Microsoft vs CrowdStrike](/compare/microsoft-vs-crowdstrike/): Microsoft 75.0 to CrowdStrike 66.0 on the published weights - [Microsoft vs SentinelOne](/compare/microsoft-vs-sentinelone/): Microsoft 75.0 to SentinelOne 62.0 on the published weights **About Bay.** Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). In regulated data, Bay ranks 3rd of 16 at 70.5. In staff rollout, Bay ranks 2nd of 16 at 73.5. Bay is a young company that left stealth in 2026, publishes no price and has little independent testing behind its claims. More: [all 16 platform profiles](/platforms/), [every head-to-head](/compare/), [the six agent surfaces](/surfaces/) and [the six agent actions](/actions/). ## Sources - learn.microsoft.com/en-us/defender-endpoint/local-agent-discovery-overview: https://learn.microsoft.com/en-us/defender-endpoint/local-agent-discovery-overview (read 12 Nov 2026) - learn.microsoft.com/en-us/defender-endpoint/ai-agent-support-matrix: https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-support-matrix (read 12 Nov 2026) - learn.microsoft.com/en-us/defender-endpoint/whats-new-in-microsoft-defender-endpoint: https://learn.microsoft.com/en-us/defender-endpoint/whats-new-in-microsoft-defender-endpoint (read 12 Nov 2026) - venturebeat.com/technology/microsoft-takes-agent-365-out-of-preview-as-shadow-ai-becomes-an-enterprise-threat: https://venturebeat.com/technology/microsoft-takes-agent-365-out-of-preview-as-shadow-ai-becomes-an-enterprise-threat (read 12 Nov 2026) - learn.microsoft.com/en-us/defender-endpoint/ai-agent-runtime-protection-overview: https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-runtime-protection-overview (read 12 Nov 2026) - www.microsoft.com/en-us/security/blog/2026/11/12/trained-to-betray-how-post-training-supply-chain-attacks-hijack-coding-agents: https://www.microsoft.com/en-us/security/blog/2026/11/12/trained-to-betray-how-post-training-supply-chain-attacks-hijack-coding-agents/ (read 13 Nov 2026) - learn.microsoft.com/en-us/defender-endpoint/configure-ai-agent-runtime-protection: https://learn.microsoft.com/en-us/defender-endpoint/configure-ai-agent-runtime-protection (read 12 Nov 2026) - learn.microsoft.com/en-us/purview/endpoint-dlp-learn-about: https://learn.microsoft.com/en-us/purview/endpoint-dlp-learn-about (read 12 Nov 2026) - learn.microsoft.com/en-us/purview/ai-microsoft-purview: https://learn.microsoft.com/en-us/purview/ai-microsoft-purview (read 12 Nov 2026) - learn.microsoft.com/en-us/entra/agent-id/what-is-microsoft-entra-agent-id: https://learn.microsoft.com/en-us/entra/agent-id/what-is-microsoft-entra-agent-id (read 12 Nov 2026) - learn.microsoft.com/en-us/microsoft-agent-365/third-party-agents: https://learn.microsoft.com/en-us/microsoft-agent-365/third-party-agents (read 12 Nov 2026) - www.microsoft.com/en-us/security/blog/2026/05/01/microsoft-agent-365-now-generally-available-expands-capabilities-and-integrations: https://www.microsoft.com/en-us/security/blog/2026/05/01/microsoft-agent-365-now-generally-available-expands-capabilities-and-integrations/ (read 12 Nov 2026) - learn.microsoft.com/en-us/purview/dspm-for-ai: https://learn.microsoft.com/en-us/purview/dspm-for-ai (read 12 Nov 2026) - learn.microsoft.com/en-us/purview/dlp-learn-about-dlp: https://learn.microsoft.com/en-us/purview/dlp-learn-about-dlp (read 12 Nov 2026) - learn.microsoft.com/en-us/defender-endpoint/microsoft-defender-endpoint: https://learn.microsoft.com/en-us/defender-endpoint/microsoft-defender-endpoint (read 12 Nov 2026) - learn.microsoft.com/en-us/microsoft-agent-365/overview: https://learn.microsoft.com/en-us/microsoft-agent-365/overview (read 12 Nov 2026) - www.microsoft.com/en-us/microsoft-365/enterprise/microsoft365-plans-and-pricing: https://www.microsoft.com/en-us/microsoft-365/enterprise/microsoft365-plans-and-pricing (read 12 Nov 2026) - www.microsoft.com/en-us/microsoft-agent-365: https://www.microsoft.com/en-us/microsoft-agent-365 (read 12 Nov 2026) - news.google.com/rss/search?q=%22Agent+365%22+price+OR+pricing+per+user&hl=en-US&gl=US&ceid=US:en (checked, nothing found): https://news.google.com/rss/search?q=%22Agent+365%22+price+OR+pricing+per+user&hl=en-US&gl=US&ceid=US:en (read 13 Nov 2026) - petri.com/microsoft-agent-365-discover-govern-ai-agents: https://petri.com/microsoft-agent-365-discover-govern-ai-agents/ (read 12 Nov 2026) - www.microsoft.com/en-us/security/business/analyst-reports (checked, nothing found): https://www.microsoft.com/en-us/security/business/analyst-reports (read 12 Nov 2026) - learn.microsoft.com/en-us/microsoft-365/admin/manage/agent-shadow-ai: https://learn.microsoft.com/en-us/microsoft-365/admin/manage/agent-shadow-ai (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/platforms/microsoft/ --- # Bay: 2nd of 16 at 71.5, strongest on Deployment at 9 of 10, the highest in the index PLATFORM PROFILE, read 13 Nov 2026 **The short answer.** Bay ranks 2nd of 16 AI endpoint security tools at 71.5 of 100. Bay and Harmonic Security tie at 71.5 and both score 8 on Sees, the heaviest criterion here, so name order puts Bay 2nd. Bay stands out on Deployment at 9 of 10, the highest in the index, and its lowest score is Transparency at 1 of 10. ## Profile - **Owner:** Bay AI, Inc. - **Product:** Bay AI agent security platform (endpoint, browser, cloud workspaces) - **Status:** Active under this name, checked 13 Nov 2026 - **Category:** AI agent security platform: discovery, posture and runtime enforcement across endpoints, browsers and cloud workspaces - **Deployment:** Agentless: one script via existing MDM or EDR using Wave, an ephemeral binary; without proxy, kernel driver or persistent agent - **Operating systems:** Windows, macOS, Linux - **Time to deploy:** under 5 minutes (vendor); live in 5 minutes, tailored in an hour - **Price:** none found on 12 Nov 2026 - **Website:** [bay.io](https://bay.io/?utm_source=aiendpointsecurityranked.com&utm_medium=referral&utm_campaign=platform) Scores out of 10 (weight)Sees 258Stops 258Deployment 159Beyond 107Policy 106Transparency 101Proof 58 ## Strength and weakness **Strength:** Bay documents on-device, session-aware allow/ask/deny/steer enforcement with named action types, agentless MDM/EDR deployment in under 5 minutes (Bay's claim), and one policy across endpoint, browser and cloud. **Where Bay is behind.** Bay is a young company that left stealth in 2026, publishes no price and has little independent testing behind its claims. Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). ## Scores: 71.5 of 100, 2nd of 16 | Criterion | Weight | Score | Why this score | | --- | --- | --- | --- | | Sees agentic activity on the endpoint | 25 | 8 | Analyst and press coverage describe an inventory of agents, browser and IDE extensions, MCP servers, skills, plugins, models and packages. The analyst notes the breadth claims are vendor-reported. | | Stops the action | 25 | 8 | Analyst and press coverage report on-device allow, ask or deny before the action for files, network, credentials, MCP calls and shell. Prevention needs the on-device component, and latency and false-positive figures are vendor-reported with no test cited. | | Deployment friction | 15 | 9 | Third-party coverage reports agentless deployment through existing MDM or EDR on Windows, Mac and Linux in under five minutes. The deploy time is the vendor's own figure and has not been independently tested. | | Beyond the endpoint | 10 | 7 | The vendor says one policy set covers devices, browsers, cloud workspaces and SaaS, and trade press restates coverage of devices, cloud and SaaS. The restatement names no browser or same-policy detail, so the detail rests on vendor claim. | | Policy and governance | 10 | 6 | The vendor offers policies scoped by IdP group, OCSF events to SIEM and session records, and trade press says it integrates with security workflows. A named EDR integration and any independent confirmation of named integrations were none found on 13 Nov 2026. | | Transparency | 10 | 1 | The only public writing is a founder anecdote. Docs are sign-in only, and no price or published research was found on 12 Nov 2026. | | Proof | 5 | 8 | Third-party press confirms $50 million raised and names customers including Micron and Peloton, and an analyst firm profiles the vendor. Trade press notes one named customer is also an investor and that the vendor's Fortune 100 usage figure is unnamed and unverified. | Bay: weighted total 71.5 of 100 on the published weights. ## What does Bay see on an endpoint? Bay documents 4 of 6 agent surfaces in full (IDE assistants, command-line agents, MCP servers and browser extensions) and 2 in part (desktop apps and SaaS copilots). Each cell with its note and source: [Bay on the Agent Surface Map](/agent-surface-map/bay/). ## Which agent actions can Bay stop? Bay documents stopping 3 of 6 agent actions before they complete (shell command, network call and credential access) and 3 in part (file read, data paste and package install). Each cell with its note and source: [Bay on the Action Control Matrix](/action-control-matrix/bay/). ## How much does Bay cost? Bay publishes no list price. None found on 12 Nov 2026. Price counts toward no score; deployment friction is scored under Deployment. ## Evidence behind the scores - Bay says its endpoint product finds every agent, extension, CLI tool, MCP server, skill and model on employee devices. (the vendor's own claim; [bay.io](https://www.bay.io/platform/ai-endpoint-security?utm_source=aiendpointsecurityranked.com&utm_medium=referral&utm_campaign=platform), read 12 Nov 2026) - Bay says the first scan maps every agent, MCP server, extension, model and skill on devices, including unapproved ones. (the vendor's own claim; [bay.io](https://www.bay.io/platform?utm_source=aiendpointsecurityranked.com&utm_medium=referral&utm_campaign=platform), read 12 Nov 2026) - Bay names Claude Code and Cursor as coding agents it secures on the endpoint, and its demo shows a Claude Code install with MCP connections. (the vendor's own claim; [bay.io](https://www.bay.io/platform/ai-endpoint-security?utm_source=aiendpointsecurityranked.com&utm_medium=referral&utm_campaign=platform), read 12 Nov 2026) - Software Analyst Cyber Research describes Bay as building an entity graph of agents, browser and IDE extensions, MCP servers, skills, plugins, models and packages across the fleet. (third-party source; [softwareanalyst.substack.com](https://softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control), read 12 Nov 2026) - SiliconANGLE reports that a Bay scanner called Wave inventories agents, extensions and local models on each machine and leaves nothing running afterward. (third-party source; [siliconangle.com](https://siliconangle.com/2026/11/10/ai-security-startup-bay-launches-with-50m-to-vet-agent-actions-before-they-run/), read 12 Nov 2026) - Software Analyst notes that breadth claims such as scanning of skills, plugins and models are vendor-reported and would benefit from validation in a proof of value. (third-party source; [softwareanalyst.substack.com](https://softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control), read 12 Nov 2026) - Bay says runtime enforcement is inline with the agentic action and returns allow, ask, steer or block in milliseconds. (the vendor's own claim; [bay.io](https://www.bay.io/platform?utm_source=aiendpointsecurityranked.com&utm_medium=referral&utm_campaign=platform), read 12 Nov 2026) - Bay says that when an agent reaches for credentials, runs a risky shell command or connects to an untrusted MCP server, it allows, asks, denies or steers the action. (the vendor's own claim; [bay.io](https://www.bay.io/platform/ai-endpoint-security?utm_source=aiendpointsecurityranked.com&utm_medium=referral&utm_campaign=platform), read 12 Nov 2026) - Software Analyst reports Bay enforces Allow/Ask/Deny on the device across files, network destinations, credential access, MCP tool calls and shell capabilities, with sub-4ms p95 latency as reported by Bay. (third-party source; [softwareanalyst.substack.com](https://softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control), read 12 Nov 2026) - Software Analyst says that once a session touches sensitive data or credentials, Bay closes outbound exfiltration routes for the rest of that session, according to Bay. (third-party source; [softwareanalyst.substack.com](https://softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control), read 12 Nov 2026) - SiliconANGLE reports that the AI tool hands each proposed action to Bay Enforcer through a native hook, on the machine, and a verdict returns in milliseconds (Bay's account). (third-party source; [siliconangle.com](https://siliconangle.com/2026/11/10/ai-security-startup-bay-launches-with-50m-to-vet-agent-actions-before-they-run/), read 12 Nov 2026) - Software Analyst cautions that Bay prevention depends on its on-device enforcement component and a collector-only deployment runs in detection mode. (third-party source; [softwareanalyst.substack.com](https://softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control), read 12 Nov 2026) - TNW reports Bay checks each proposed agent action against the whole session and decides locally, in milliseconds, whether it can go ahead (Bay's account). (third-party source; [thenextweb.com](https://thenextweb.com/news/bay-50m-ai-agent-security-stealth-battery-ventures), read 13 Nov 2026) - Unite.AI says Bay's central challenge remains showing that session-level enforcement can support wider agent autonomy without multiplying the rules security teams must maintain. (third-party source; [unite.ai](https://www.unite.ai/bay-raises-37m-series-a-to-scale-enterprise-ai-agent-security/), read 13 Nov 2026) ## Checks that found nothing - Third-party or observed confirmation of named SIEM, Splunk, Okta or EDR integrations for Bay; news searches (query C5) returned only the 10 Nov 2026 funding coverage and no docs are public (docs.bay.io requires sign-in): none found on 13 Nov 2026 (checked: [docs.bay.io](https://docs.bay.io/?utm_source=aiendpointsecurityranked.com&utm_medium=referral&utm_campaign=platform)) - A published price. bay.io/pricing returns 404 and the sitemap lists no pricing page; calls to action are "Get a demo": none found on 12 Nov 2026 (checked: [bay.io](https://www.bay.io/sitemap.xml?utm_source=aiendpointsecurityranked.com&utm_medium=referral&utm_campaign=platform)) - Published research, CVE notices or release notes on bay.io; the blog lists only two launch posts dated November 10, 2026: none found on 12 Nov 2026 (checked: [bay.io](https://www.bay.io/blog?utm_source=aiendpointsecurityranked.com&utm_medium=referral&utm_campaign=platform)) ## What Bay says about itself These are Bay's own statements, each confirmed on the page named and read 12 Nov 2026. - Positioning: "The AI Agent Security Platform." ([bay.io](https://www.bay.io/?utm_source=aiendpointsecurityranked.com&utm_medium=referral&utm_campaign=platform), read 12 Nov 2026) - "Calm the chaos. Bay is the single control plane for all agentic activity, wherever it runs." ([bay.io](https://www.bay.io/?utm_source=aiendpointsecurityranked.com&utm_medium=referral&utm_campaign=platform), read 12 Nov 2026) - Meta description: single control plane for agentic activity on endpoints, browsers and cloud workspaces... ([bay.io](https://www.bay.io/?utm_source=aiendpointsecurityranked.com&utm_medium=referral&utm_campaign=platform), read 12 Nov 2026) - Coverage: "Every agent, MCP server, plugin and skill, wherever it runs", on laptops, browsers and cloud workspaces ([bay.io](https://www.bay.io/?utm_source=aiendpointsecurityranked.com&utm_medium=referral&utm_campaign=platform), read 12 Nov 2026) - Deployment: agentless, "Live in under 5 minutes, on any surface" ([bay.io](https://www.bay.io/?utm_source=aiendpointsecurityranked.com&utm_medium=referral&utm_campaign=platform), read 12 Nov 2026) - Control: "Risky actions stop before they complete" ([bay.io](https://www.bay.io/?utm_source=aiendpointsecurityranked.com&utm_medium=referral&utm_campaign=platform), read 12 Nov 2026) - Guardrails calibrated to what the agent is doing ([bay.io](https://www.bay.io/?utm_source=aiendpointsecurityranked.com&utm_medium=referral&utm_campaign=platform), read 12 Nov 2026) - "customized out-of-the-box policies" ([bay.io](https://www.bay.io/?utm_source=aiendpointsecurityranked.com&utm_medium=referral&utm_campaign=platform), read 12 Nov 2026) ## Position in each situation | Situation | Position | Total | Leader | | --- | --- | --- | --- | | [Coding agents](/best-for/coding-agents/) | 1st of 16 (tied) | 75.5 | Bay 75.5 | | [Regulated data](/best-for/regulated-data/) | 3rd of 16 | 70.5 | Microsoft 79.0 | | [Lean team with EDR](/best-for/lean-team-edr/) | 1st of 16 | 76.5 | Bay 76.5 | | [Staff rollout](/best-for/staff-rollout/) | 2nd of 16 | 73.5 | Microsoft 75.5 | | [Board in 30 days](/best-for/board-30-days/) | 1st of 16 | 77.5 | Bay 77.5 | Bay under each situation's own weights. Tied marks an equal total; the order follows the tie-break, the higher score on the heaviest criterion, then name. ## Head-to-heads - [Bay vs Island](/compare/bay-vs-island/): Bay 71.5 to Island 66.0 on the published weights - [Bay vs Microsoft](/compare/bay-vs-microsoft/): Microsoft 75.0 to Bay 71.5 on the published weights - [Bay vs Knostic](/compare/bay-vs-knostic/): Bay 71.5 to Knostic 64.0 on the published weights - [Bay vs WitnessAI](/compare/bay-vs-witnessai/): Bay 71.5 to WitnessAI 61.0 on the published weights - [Bay vs Pillar Security](/compare/bay-vs-pillar-security/): Bay 71.5 to Pillar Security 53.0 on the published weights - [Bay vs Aim Security](/compare/bay-vs-aim-security/): Bay 71.5 to Aim Security 69.0 on the published weights ## Where Bay stands In regulated data, Bay ranks 3rd of 16 at 70.5, behind Microsoft at 79.0. Microsoft's lead over Bay comes from policy and governance (9 to 6) and transparency (7 to 1). In staff rollout, Bay ranks 2nd of 16 at 73.5, behind Microsoft at 75.5. Microsoft's lead over Bay comes from policy and governance (9 to 6) and transparency (7 to 1). More: [all 16 platform profiles](/platforms/), [every head-to-head](/compare/), [the six agent surfaces](/surfaces/) and [the six agent actions](/actions/). ## Sources - www.bay.io/platform/ai-endpoint-security: https://www.bay.io/platform/ai-endpoint-security (read 12 Nov 2026) - www.bay.io/platform: https://www.bay.io/platform (read 12 Nov 2026) - softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control: https://softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control (read 12 Nov 2026) - siliconangle.com/2026/11/10/ai-security-startup-bay-launches-with-50m-to-vet-agent-actions-before-they-run: https://siliconangle.com/2026/11/10/ai-security-startup-bay-launches-with-50m-to-vet-agent-actions-before-they-run/ (read 12 Nov 2026) - thenextweb.com/news/bay-50m-ai-agent-security-stealth-battery-ventures: https://thenextweb.com/news/bay-50m-ai-agent-security-stealth-battery-ventures (read 13 Nov 2026) - www.unite.ai/bay-raises-37m-series-a-to-scale-enterprise-ai-agent-security: https://www.unite.ai/bay-raises-37m-series-a-to-scale-enterprise-ai-agent-security/ (read 13 Nov 2026) - www.scworld.com/brief/bay-ai-launches-with-50-million-to-secure-enterprise-ai-agents: https://www.scworld.com/brief/bay-ai-launches-with-50-million-to-secure-enterprise-ai-agents (read 13 Nov 2026) - www.bay.io: https://www.bay.io (read 12 Nov 2026) - www.securityweek.com/bay-raises-50-million-to-control-what-enterprise-ai-agents-can-do: https://www.securityweek.com/bay-raises-50-million-to-control-what-enterprise-ai-agents-can-do/ (read 12 Nov 2026) - www.bay.io/platform/cloud-workspaces: https://www.bay.io/platform/cloud-workspaces (read 12 Nov 2026) - www.bay.io/platform/browser: https://www.bay.io/platform/browser (read 12 Nov 2026) - www.bay.io/blog/bay-emerges-from-stealth-with-50m-after-fortune-100-companies-used-its-platform-to-triple-autonomous-ai-agent-use-2: https://www.bay.io/blog/bay-emerges-from-stealth-with-50m-after-fortune-100-companies-used-its-platform-to-triple-autonomous-ai-agent-use-2 (read 12 Nov 2026) - pulse2.com/bay-50-million-funding-to-expand-ai-agent-security-platform-across-enterprise-environments: https://pulse2.com/bay-50-million-funding-to-expand-ai-agent-security-platform-across-enterprise-environments/ (read 13 Nov 2026) - www.bay.io/ai-agent-investigation: https://www.bay.io/ai-agent-investigation (read 12 Nov 2026) - www.bay.io/ai-governance-compliance: https://www.bay.io/ai-governance-compliance (read 12 Nov 2026) - docs.bay.io (checked, nothing found): https://docs.bay.io (read 13 Nov 2026) - www.bay.io/blog/introducing-bay-the-ai-agent-security-platform-out-of-stealth-with-50m: https://www.bay.io/blog/introducing-bay-the-ai-agent-security-platform-out-of-stealth-with-50m (read 12 Nov 2026) - www.bay.io/sitemap.xml (checked, nothing found): https://www.bay.io/sitemap.xml (read 12 Nov 2026) - www.bay.io/blog (checked, nothing found): https://www.bay.io/blog (read 12 Nov 2026) - www.calcalistech.com/ctechnews/article/bjx11v5x0gx: https://www.calcalistech.com/ctechnews/article/bjx11v5x0gx (read 12 Nov 2026) - docs.harmonicsecurity.app/portal-guides/integration-guides/configure-opentelemetry: https://docs.harmonicsecurity.app/portal-guides/integration-guides/configure-opentelemetry (read 12 Nov 2026) - docs.harmonicsecurity.app/portal-guides/integration-guides/configure-anthropic-compliance-api-connector: https://docs.harmonicsecurity.app/portal-guides/integration-guides/configure-anthropic-compliance-api-connector (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/platforms/bay/ --- # Harmonic Security: 3rd of 16 at 71.5, strongest on Sees at 8 of 10 PLATFORM PROFILE, read 13 Nov 2026 **The short answer.** Harmonic Security ranks 3rd of 16 AI endpoint security tools at 71.5 of 100. Bay and Harmonic Security tie at 71.5 and both score 8 on Sees, the heaviest criterion here, so name order puts Bay 2nd. Harmonic Security stands out on Sees at 8 of 10, and its lowest score is Transparency at 6 of 10. ## Profile - **Owner:** Harmonic Security Inc. (independent; no parent or acquirer named on its pages) - **Product:** Harmonic Protect Agent, MCP Gateway and Command (AI Governance and Control platform) - **Status:** Active under this name, checked 13 Nov 2026 - **Category:** Workforce AI usage control: endpoint agent, browser extension and local MCP gateway - **Deployment:** Endpoint agent (MSI/DEB/RPM/PKG via MDM) plus browser extension plus local MCP gateway; no network proxy - **Operating systems:** Windows 11, macOS 15 to 27, Ubuntu 22.04/24.04/26.04 LTS, Rocky Linux 9 and 10 - **Time to deploy:** none found on 12 Nov 2026 - **Price:** none found on 12 Nov 2026 (three tiers shown, no dollar amounts) - **Website:** [harmonic.security](https://www.harmonic.security/) Scores out of 10 (weight)Sees 258Stops 257Deployment 157Beyond 107Policy 107Transparency 106Proof 57 ## Strength and weakness **Strength:** Harmonic documents hook-based inline blocking of prompts and tool calls across named coding agents (Claude Code, Codex, Cursor, Copilot CLI and others) plus an MCP gateway with identity-aware access groups. **Weakness:** Desktop-app controls are Windows-only, prompt-only and early access in the docs, the Cloud MCP Gateway does not connect every remote MCP server, and none found on 13 Nov 2026 for blocking file reads, shell commands or package installs as named action types. ## Scores: 71.5 of 100, 3rd of 16 | Criterion | Weight | Score | Why this score | | --- | --- | --- | --- | | Sees agentic activity on the endpoint | 25 | 8 | The vendor names desktop apps, AI IDEs, command-line agents, local models and MCP servers, and observed docs show OTel ingestion and Compliance API visibility. No published inventory of skills or plugins was found on 12 Nov 2026, and coverage is set per customer. | | Stops the action | 25 | 7 | Observed docs and release notes show hook-based blocking of prompts and tool calls for many named coding agents, plus MCP gateway blocks, with limits written down, including that uploaded files are not covered. Separate blocked action types such as file reads, shell commands or credential access were none found on 12 Nov 2026, and an independent test was none found on 13 Nov 2026. | | Deployment friction | 15 | 7 | Observed docs list supported Windows, macOS and Linux versions and MDM installers with no proxy. A published time-to-deploy figure was none found on 12 Nov 2026, and an independent deployment source was none found on 13 Nov 2026. | | Beyond the endpoint | 10 | 7 | The same platform covers browser tools, desktop apps, MCP agents and cloud-to-cloud agent connections, with cloud and SaaS paths observed in docs and release notes. The cloud piece is early access, and its docs say not every remote MCP server can be connected, naming six that do not work. | | Policy and governance | 10 | 7 | Observed docs show identity-based access groups, Okta enrichment and webhook event types for SIEM; a full audit log is an add-on. A named EDR integration was none found on 12 Nov 2026, and an independent source on integrations was none found on 13 Nov 2026. | | Transparency | 10 | 6 | Public documentation and dated release notes were observed, and the vendor publishes usage research. No dollar price and no agent-vulnerability research were found on 12 Nov 2026. | | Proof | 5 | 7 | Third-party press confirms $50 million Series B funding and a regional distribution deal, the vendor names two customers, and it says Gartner lists it as an Example Vendor. The customer names and analyst mention are vendor-published only. | Harmonic Security: weighted total 71.5 of 100 on the published weights. ## What does Harmonic Security see on an endpoint? Harmonic Security documents 5 of 6 agent surfaces in full (desktop apps, IDE assistants, command-line agents, MCP servers and browser extensions) and 1 in part (SaaS copilots). Each cell with its note and source: [Harmonic Security on the Agent Surface Map](/agent-surface-map/harmonic-security/). ## Which agent actions can Harmonic Security stop? Harmonic Security documents stopping 1 of 6 agent actions before they complete (data paste) and 3 in part (shell command, network call and credential access). Nothing was found for file read and package install. Each cell with its note and source: [Harmonic Security on the Action Control Matrix](/action-control-matrix/harmonic-security/). ## How much does Harmonic Security cost? Harmonic Security publishes no list price. None found on 12 Nov 2026 (three tiers shown, no dollar amounts) Price counts toward no score; deployment friction is scored under Deployment. ## Evidence behind the scores - Harmonic says its Endpoint Agent identifies native AI apps, AI-first IDEs, CLI tools, local models and direct inference API calls on employee devices. (the vendor's own claim; [harmonic.security](https://www.harmonic.security/solutions/endpoint-ai-security), read 12 Nov 2026) - Harmonic names Claude Desktop, ChatGPT Desktop, Cursor, Claude Code, GitHub Copilot, Ollama and MCP servers on developer machines as covered by app discovery. (the vendor's own claim; [harmonic.security](https://www.harmonic.security/solutions/endpoint-ai-security), read 12 Nov 2026) - Harmonic's Command tier adds MCP client and server discovery and monitoring of what agents are doing and which systems they access. (the vendor's own claim; [harmonic.security](https://www.harmonic.security/pricing), read 12 Nov 2026) - Harmonic's OpenTelemetry guide says it ingests prompt and tool visibility from any AI application that can emit OTel logs, with Claude Code and Codex configured from the Harmonic portal by the Protect Agent. (observed in public documentation or code; [docs.harmonicsecurity.app](https://docs.harmonicsecurity.app/portal-guides/integration-guides/configure-opentelemetry), read 12 Nov 2026) - Harmonic's connector guide says it reads Claude Enterprise activity (chats, projects, files, users, audit activities) through the Anthropic Compliance API, giving prompt-level visibility for Claude.ai without the browser extension or endpoint agent. (observed in public documentation or code; [docs.harmonicsecurity.app](https://docs.harmonicsecurity.app/portal-guides/integration-guides/configure-anthropic-compliance-api-connector), read 12 Nov 2026) - Harmonic's Protect Agent FAQ lists real-time reporting of desktop AI usage, including connections to known AI network hosts, as one of the agent's configurable features. (observed in public documentation or code; [docs.harmonicsecurity.app](https://docs.harmonicsecurity.app/protect-agent/installation-and-deployment/frequently-asked-questions), read 12 Nov 2026) - Harmonic's documentation says coding-agent inline controls use each application's built-in hooks and are supported on every OS the Protect Agent supports. (observed in public documentation or code; [docs.harmonicsecurity.app](https://docs.harmonicsecurity.app/portal-guides/detection-and-response/sensitive-data-controls-for-coding-agents-and-desktop-applications), read 12 Nov 2026) - Harmonic documents that coding-agent inline controls cover user prompts and tool calls, for Claude Code, Codex, ChatGPT Work, Cursor, VS Code, Copilot CLI, Antigravity, Grok Build and Kiro CLI. (observed in public documentation or code; [docs.harmonicsecurity.app](https://docs.harmonicsecurity.app/portal-guides/detection-and-response/sensitive-data-controls-for-coding-agents-and-desktop-applications), read 12 Nov 2026) - Harmonic's MCP access-control docs say an attempt to use an MCP server a person lacks permission for is blocked by the gateway and creates an access request for an administrator. (observed in public documentation or code; [docs.harmonicsecurity.app](https://docs.harmonicsecurity.app/portal-guides/managing-your-organization/configure-mcp-server-access-controls), read 12 Nov 2026) - Harmonic's scenario docs say that for MCP tool calls containing sensitive data the Block intervention blocks the request and tells the agent what was detected so it can fix the request. (observed in public documentation or code; [docs.harmonicsecurity.app](https://docs.harmonicsecurity.app/portal-guides/detection-and-response/configure-scenarios-and-interventions), read 12 Nov 2026) - Harmonic's docs list limits: on Grok Build prompts can be alerted but not blocked, and Antigravity coverage is tool calls only on macOS and Linux. (observed in public documentation or code; [docs.harmonicsecurity.app](https://docs.harmonicsecurity.app/portal-guides/detection-and-response/sensitive-data-controls-for-coding-agents-and-desktop-applications), read 12 Nov 2026) - Harmonic's prompt-injection docs (early access) state a limit: coding-agent hooks cannot fully block tool-result content, so the alert is raised but the content still reaches the agent's context. (observed in public documentation or code; [docs.harmonicsecurity.app](https://docs.harmonicsecurity.app/portal-guides/detection-and-response/prompt-injection-detection-early-access), read 12 Nov 2026) - Harmonic's August 13, 2026 release note limits desktop-app and coding-agent controls to prompts and tool calls, not uploaded files. (observed in public documentation or code; [docs.harmonicsecurity.app](https://docs.harmonicsecurity.app/release-notes/2026/august-13-2026.md), read 13 Nov 2026) - Harmonic's August 19, 2026 release note says the hosted gateway blocks a remote MCP server an employee is not permitted to use and raises an Access Request (Early Access). (observed in public documentation or code; [docs.harmonicsecurity.app](https://docs.harmonicsecurity.app/release-notes/2026/august-19-2026.md), read 13 Nov 2026) ## Checks that found nothing - Harmonic documentation naming blocking of file reads, shell commands, package installs or credential-file access as separate action types; the documented hook controls cover prompts and tool calls: none found on 12 Nov 2026 (checked: [docs.harmonicsecurity.app](https://docs.harmonicsecurity.app/portal-guides/detection-and-response/sensitive-data-controls-for-coding-agents-and-desktop-applications)) - An independent test, analyst write-up or trade-press article confirming Harmonic blocks agent actions before completion; Google News queries returned only funding, channel and survey coverage: none found on 13 Nov 2026 (checked: [news.google.com](https://news.google.com/rss/search?hl=en-US&gl=US&ceid=US:en&q=%22Harmonic+Security%22+block+OR+runtime+OR+enforcement)) - A published time-to-deploy figure for the Protect Agent or MCP Gateway on the pricing page, product pages or deployment FAQ: none found on 12 Nov 2026 (checked: [harmonic.security](https://www.harmonic.security/pricing)) - A third-party or independent source on Harmonic Protect Agent deployment, operating systems or time to deploy; Google News query returned funding and survey coverage only: none found on 13 Nov 2026 (checked: [news.google.com](https://news.google.com/rss/search?hl=en-US&gl=US&ceid=US:en&q=%22Harmonic+Security%22+agentless+OR+deploy+OR+macOS+OR+Windows)) - A third-party or independent source on Harmonic SIEM, Okta or audit-log integrations; Google News query returned funding and survey coverage only: none found on 13 Nov 2026 (checked: [news.google.com](https://news.google.com/rss/search?hl=en-US&gl=US&ceid=US:en&q=%22Harmonic+Security%22+SIEM+OR+Splunk+OR+Okta+OR+audit)) - Any published dollar price or per-seat figure for Explore, Guide or Command on the pricing page: none found on 12 Nov 2026 (checked: [harmonic.security](https://www.harmonic.security/pricing)) - CVE notices or agent-vulnerability research by Harmonic in its resource library listing: none found on 12 Nov 2026 (checked: [harmonic.security](https://www.harmonic.security/resources)) ## Position in each situation | Situation | Position | Total | Leader | | --- | --- | --- | --- | | [Coding agents](/best-for/coding-agents/) | 3rd of 16 | 72.0 | Bay 75.5 | | [Regulated data](/best-for/regulated-data/) | 2nd of 16 | 72.5 | Microsoft 79.0 | | [Lean team with EDR](/best-for/lean-team-edr/) | 3rd of 16 | 72.5 | Bay 76.5 | | [Staff rollout](/best-for/staff-rollout/) | 3rd of 16 | 71.5 | Microsoft 75.5 | | [Board in 30 days](/best-for/board-30-days/) | 2nd of 16 | 73.5 | Bay 77.5 | Harmonic Security under each situation's own weights. Tied marks an equal total; the order follows the tie-break, the higher score on the heaviest criterion, then name. ## Head-to-heads - [Knostic vs Harmonic Security](/compare/knostic-vs-harmonic-security/): Harmonic Security 71.5 to Knostic 64.0 on the published weights **About Bay.** Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). In regulated data, Bay ranks 3rd of 16 at 70.5. In staff rollout, Bay ranks 2nd of 16 at 73.5. Bay is a young company that left stealth in 2026, publishes no price and has little independent testing behind its claims. More: [all 16 platform profiles](/platforms/), [every head-to-head](/compare/), [the six agent surfaces](/surfaces/) and [the six agent actions](/actions/). ## Sources - www.harmonic.security/solutions/endpoint-ai-security: https://www.harmonic.security/solutions/endpoint-ai-security (read 12 Nov 2026) - www.harmonic.security/pricing: https://www.harmonic.security/pricing (read 12 Nov 2026) - docs.harmonicsecurity.app/portal-guides/integration-guides/configure-opentelemetry: https://docs.harmonicsecurity.app/portal-guides/integration-guides/configure-opentelemetry (read 12 Nov 2026) - docs.harmonicsecurity.app/portal-guides/integration-guides/configure-anthropic-compliance-api-connector: https://docs.harmonicsecurity.app/portal-guides/integration-guides/configure-anthropic-compliance-api-connector (read 12 Nov 2026) - docs.harmonicsecurity.app/protect-agent/installation-and-deployment/frequently-asked-questions: https://docs.harmonicsecurity.app/protect-agent/installation-and-deployment/frequently-asked-questions (read 12 Nov 2026) - docs.harmonicsecurity.app/portal-guides/detection-and-response/sensitive-data-controls-for-coding-agents-and-desktop-applications: https://docs.harmonicsecurity.app/portal-guides/detection-and-response/sensitive-data-controls-for-coding-agents-and-desktop-applications (read 12 Nov 2026) - docs.harmonicsecurity.app/portal-guides/managing-your-organization/configure-mcp-server-access-controls: https://docs.harmonicsecurity.app/portal-guides/managing-your-organization/configure-mcp-server-access-controls (read 12 Nov 2026) - docs.harmonicsecurity.app/portal-guides/detection-and-response/configure-scenarios-and-interventions: https://docs.harmonicsecurity.app/portal-guides/detection-and-response/configure-scenarios-and-interventions (read 12 Nov 2026) - docs.harmonicsecurity.app/portal-guides/detection-and-response/prompt-injection-detection-early-access: https://docs.harmonicsecurity.app/portal-guides/detection-and-response/prompt-injection-detection-early-access (read 12 Nov 2026) - docs.harmonicsecurity.app/release-notes/2026/august-13-2026.md: https://docs.harmonicsecurity.app/release-notes/2026/august-13-2026.md (read 13 Nov 2026) - docs.harmonicsecurity.app/release-notes/2026/august-19-2026.md: https://docs.harmonicsecurity.app/release-notes/2026/august-19-2026.md (read 13 Nov 2026) - news.google.com/rss/search?hl=en-US&gl=US&ceid=US:en&q=%22Harmonic+Security%22+block+OR+runtime+OR+enforcement (checked, nothing found): https://news.google.com/rss/search?hl=en-US&gl=US&ceid=US:en&q=%22Harmonic+Security%22+block+OR+runtime+OR+enforcement (read 13 Nov 2026) - www.harmonic.security/products/command: https://www.harmonic.security/products/command (read 12 Nov 2026) - docs.harmonicsecurity.app/protect-agent/installation-and-deployment/linux: https://docs.harmonicsecurity.app/protect-agent/installation-and-deployment/linux (read 12 Nov 2026) - docs.harmonicsecurity.app/portal-guides/managing-your-organization/release-groups.md: https://docs.harmonicsecurity.app/portal-guides/managing-your-organization/release-groups.md (read 13 Nov 2026) - news.google.com/rss/search?hl=en-US&gl=US&ceid=US:en&q=%22Harmonic+Security%22+agentless+OR+deploy+OR+macOS+OR+Windows (checked, nothing found): https://news.google.com/rss/search?hl=en-US&gl=US&ceid=US:en&q=%22Harmonic+Security%22+agentless+OR+deploy+OR+macOS+OR+Windows (read 13 Nov 2026) - docs.harmonicsecurity.app/mcp-gateway/getting-started/what-is-mcp-gateway: https://docs.harmonicsecurity.app/mcp-gateway/getting-started/what-is-mcp-gateway (read 12 Nov 2026) - docs.harmonicsecurity.app/llms.txt: https://docs.harmonicsecurity.app/llms.txt (read 12 Nov 2026) - docs.harmonicsecurity.app/mcp-gateway/cloud-mcp-gateway/creating-a-new-connector.md: https://docs.harmonicsecurity.app/mcp-gateway/cloud-mcp-gateway/creating-a-new-connector.md (read 13 Nov 2026) - docs.harmonicsecurity.app/portal-guides/integration-guides/okta-integration: https://docs.harmonicsecurity.app/portal-guides/integration-guides/okta-integration (read 12 Nov 2026) - docs.harmonicsecurity.app/portal-guides/integration-guides/configure-a-webhook: https://docs.harmonicsecurity.app/portal-guides/integration-guides/configure-a-webhook (read 12 Nov 2026) - www.harmonic.security: https://www.harmonic.security (read 12 Nov 2026) - news.google.com/rss/search?hl=en-US&gl=US&ceid=US:en&q=%22Harmonic+Security%22+SIEM+OR+Splunk+OR+Okta+OR+audit (checked, nothing found): https://news.google.com/rss/search?hl=en-US&gl=US&ceid=US:en&q=%22Harmonic+Security%22+SIEM+OR+Splunk+OR+Okta+OR+audit (read 13 Nov 2026) - docs.harmonicsecurity.app: https://docs.harmonicsecurity.app/ (read 12 Nov 2026) - docs.harmonicsecurity.app/release-notes/2026/september-25-2026: https://docs.harmonicsecurity.app/release-notes/2026/september-25-2026 (read 12 Nov 2026) - www.harmonic.security/resources: https://www.harmonic.security/resources (read 12 Nov 2026) - siliconangle.com/2026/10/14/harmonic-security-raises-50m-to-govern-how-employees-use-ai-tools: https://siliconangle.com/2026/10/14/harmonic-security-raises-50m-to-govern-how-employees-use-ai-tools/ (read 12 Nov 2026) - www.securityweek.com/harmonic-security-raises-50-million-for-workforce-ai-protection: https://www.securityweek.com/harmonic-security-raises-50-million-for-workforce-ai-protection/ (read 12 Nov 2026) - www.harmonic.security/resources/harmonic-security-launches-mcp-gateway-providing-complete-visibility-and-control-over-an-organizations-agentic-ai-ecosystem: https://www.harmonic.security/resources/harmonic-security-launches-mcp-gateway-providing-complete-visibility-and-control-over-an-organizations-agentic-ai-ecosystem (read 12 Nov 2026) - www.harmonic.security/resources/gartner-ai-usage-control-research-points-beyond-basic-discovery: https://www.harmonic.security/resources/gartner-ai-usage-control-research-points-beyond-basic-discovery (read 12 Nov 2026) - securitybrief.co.uk/story/distology-signs-harmonic-security-deal-for-ai-governance: https://securitybrief.co.uk/story/distology-signs-harmonic-security-deal-for-ai-governance (read 13 Nov 2026) - softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control: https://softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control (read 12 Nov 2026) - siliconangle.com/2026/11/10/ai-security-startup-bay-launches-with-50m-to-vet-agent-actions-before-they-run: https://siliconangle.com/2026/11/10/ai-security-startup-bay-launches-with-50m-to-vet-agent-actions-before-they-run/ (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/platforms/harmonic-security/ --- # Aim Security: 4th of 16 at 69.0, strongest on Transparency at 7 of 10 PLATFORM PROFILE, read 13 Nov 2026 **The short answer.** Aim Security ranks 4th of 16 AI endpoint security tools at 69.0 of 100. Aim Security stands out on Transparency at 7 of 10, and its lowest score is Deployment at 5 of 10. Aim Security shows 4 of 6 agent surfaces in full and can stop 1 of 6 agent actions, as documented. ## Profile - **Owner:** Cato Networks (acquired Aim Security, announced September 3, 2025) - **Product:** Cato AI Security (formerly Aim Security), including Cato AI Endpoint - **Status:** Active under this name, checked 13 Nov 2026 - **Category:** AI security inside a SASE platform: endpoint client plus inline network inspection for employee AI use and agents - **Deployment:** Endpoint client (lightweight, pushed with existing endpoint management tools) plus inline network inspection through the Cato SASE platform; agentless discovery for AI usage mapping - **Operating systems:** Windows (Cato Client v6.4 and higher, AI Only mode) - **Time to deploy:** vendor claim: 'in weeks'; no measured figure published - **Price:** none found on 12 Nov 2026 (Cato docs say AI for End Users and AI for Applications are separate per-user licenses; no figure) - **Website:** [catonetworks.com](https://www.catonetworks.com/platform/cato-ai-endpoint/) Scores out of 10 (weight)Sees 258Stops 257Deployment 155Beyond 107Policy 106Transparency 107Proof 58 ## Strength and weakness **Strength:** Cato AI Endpoint documents a lightweight client that finds local agents with their MCP servers and skills and blocks an unapproved tool call before it runs, in one dashboard with inline network inspection. **Weakness:** None found on 13 Nov 2026 for a full operating-system list (the AI Only client mode is documented for Windows Client v6.4 and higher only), a published price, or public AI Endpoint documentation; blocking is limited to 'supported local agents' and pasted data is checked at send, not at paste. ## Scores: 69.0 of 100, 4th of 16 | Criterion | Weight | Score | Why this score | | --- | --- | --- | --- | | Sees agentic activity on the endpoint | 25 | 8 | The vendor names local agents with their MCP servers and skills, plus browser and desktop AI, and observed product updates show inventory pages for agents, tools, MCP servers and skills, coding agent sessions and MCP discovery for a named agent. Discovery coverage of IDE extensions is not detailed in the observed records. | | Stops the action | 25 | 7 | Observed product updates show Local Agent policy rules with Block, Engage and Monitor actions and a detector that blocks destructive commands before they delete files or data, alongside network blocking of AI apps. Blocking is limited to supported local agents, and named blocking of credential access or network calls was none found on 13 Nov 2026. | | Deployment friction | 15 | 5 | A light client is pushed with existing tools, the vendor claims go-live in weeks, and observed product updates tie the AI traffic mode to Windows Client v6.4 and higher. A macOS or full operating-system list for the AI endpoint client was none found on 13 Nov 2026. | | Beyond the endpoint | 10 | 7 | The vendor says the same engine and rules cover the browser, network AI traffic, managed agents and cloud workloads, and observed product updates show SaaS posture checks expanding to over 20 apps. The observed piece is posture checking rather than the same enforcement policy, so the cross-surface policy rests on vendor claim. | | Policy and governance | 10 | 6 | The vendor describes allow, restrict and monitor policy with audit trails, and observed docs show policy configuration and a coding agent event type sent to third-party systems such as a SIEM. A documented EDR or identity-provider integration for AI events was none found on 13 Nov 2026. | | Transparency | 10 | 7 | Public platform docs and release notes were observed, and third-party sources confirm a CVE from its research. No published price and no AI endpoint-specific documentation page were found on 12 Nov 2026. | | Proof | 5 | 8 | Third-party press confirms funding, an acquisition and a named customer, and the vendor names further customers and a Gartner Leader placement. The analyst placement is platform-level, not specific to AI endpoints. | Aim Security: weighted total 69.0 of 100 on the published weights. ## What does Aim Security see on an endpoint? Aim Security documents 4 of 6 agent surfaces in full (desktop apps, command-line agents, MCP servers and browser extensions) and 2 in part (IDE assistants and SaaS copilots). Each cell with its note and source: [Aim Security on the Agent Surface Map](/agent-surface-map/aim-security/). ## Which agent actions can Aim Security stop? Aim Security documents stopping 1 of 6 agent actions before they complete (data paste) and 3 in part (file read, shell command and network call). Nothing was found for credential access and package install. Each cell with its note and source: [Aim Security on the Action Control Matrix](/action-control-matrix/aim-security/). ## How much does Aim Security cost? Aim Security publishes no list price. None found on 12 Nov 2026 (Cato docs say AI for End Users and AI for Applications are separate per-user licenses; no figure) Price counts toward no score; deployment friction is scored under Deployment. ## Evidence behind the scores - Cato says its AI Endpoint finds local agents plus the MCP servers and skills that extend each agent. (the vendor's own claim; [catonetworks.com](https://www.catonetworks.com/platform/cato-ai-endpoint/), read 12 Nov 2026) - Cato says a lightweight client covers browser AI, desktop AI apps and add-ins, and local agents on the endpoint. (the vendor's own claim; [catonetworks.com](https://www.catonetworks.com/platform/cato-ai-endpoint/), read 12 Nov 2026) - The Cato AI Endpoint page shows example local agents (Claude Code, Cursor, Codex) with per-user and per-endpoint risk flags such as full command-line access. (the vendor's own claim; [catonetworks.com](https://www.catonetworks.com/platform/cato-ai-endpoint/), read 12 Nov 2026) - Cato's October 2026 newswire release says the AI Endpoint client discovers AI use across browsers, desktop applications and local agents on the endpoint. (the vendor's own claim; [prnewswire.com](https://www.prnewswire.com/news-releases/cato-networks-launches-cato-ai-endpoint-to-secure-ai-use-beyond-the-browser-302919573.html), read 12 Nov 2026) - Cato's October 19, 2026 product updates describe an AI Security monitoring menu with inventory pages for users, applications, agents, tools, MCP servers and skills, and separate coding agent sessions. (observed in public documentation or code; [knowledge.catonetworks.com](https://knowledge.catonetworks.com/docs/product-updates-october-19-2026.md), read 13 Nov 2026) - Cato's November 9, 2026 product updates say AI Security now supports Antigravity controls in all tenants, with a Scout tab for discovery of its MCPs and tools. (observed in public documentation or code; [knowledge.catonetworks.com](https://knowledge.catonetworks.com/docs/product-updates-november-9-2026.md), read 13 Nov 2026) - Cato's knowledge base documents pulling Claude conversation logs and user activity through the Anthropic Compliance API, a SaaS-side view of Claude use. (observed in public documentation or code; [knowledge.catonetworks.com](https://knowledge.catonetworks.com/docs/integrating-anthropic-compliance-api-with-cato-ai-security), read 12 Nov 2026) - TechCrunch reported in January 2024 that Aim's platform includes a lightweight browser extension and knows which GenAI tools employees use and what data they feed them. (third-party source; [techcrunch.com](https://techcrunch.com/2024/01/31/aim-security-raises-10m-for-its-genai-security-platform/), read 12 Nov 2026) - Candid.Technology, an independent publisher, lists Aim Security among shadow agent monitoring tools and says it pairs discovery of AI usage with policy enforcement (a roundup, not a product test). (third-party source; [candid.technology](https://candid.technology/best-shadow-agent-monitoring-solutions-for-enterprises/), read 13 Nov 2026) - Cato says AI Endpoint blocks an unapproved tool call from a supported local agent on the endpoint before it runs. (the vendor's own claim; [catonetworks.com](https://www.catonetworks.com/platform/cato-ai-endpoint/), read 12 Nov 2026) - Cato says its agent security inspects prompts, responses, tool calls and tool actions and blocks risky actions before they execute. (the vendor's own claim; [catonetworks.com](https://www.catonetworks.com/platform/agentic-ai-security/), read 12 Nov 2026) - Cato's November 9, 2026 product updates say a Destructive Commands Detector can block or monitor commands in AI interactions before they delete files, resources or data; it needs the AI Security for Users license. (observed in public documentation or code; [knowledge.catonetworks.com](https://knowledge.catonetworks.com/docs/product-updates-november-9-2026.md), read 13 Nov 2026) - Cato's October 26, 2026 product updates say Local Agent policy rules can now use an Engage action, which verifies user intent for risky agent actions, in addition to Monitor and Block. (observed in public documentation or code; [knowledge.catonetworks.com](https://knowledge.catonetworks.com/docs/product-updates-october-26-2026.md), read 13 Nov 2026) - Cato says that for desktop AI apps it checks the prompt when sent and stops sensitive data reaching the AI service; it does not act while the user types or pastes. (the vendor's own claim; [catonetworks.com](https://www.catonetworks.com/platform/cato-ai-endpoint/), read 12 Nov 2026) ## Checks that found nothing - An operating-system list or install guide for Cato AI Endpoint on the product page or in public documentation: none found on 12 Nov 2026 (checked: [catonetworks.com](https://www.catonetworks.com/platform/cato-ai-endpoint/)) - The Cato knowledge base lists a separate EPP agent with versioned installers and manual upgrade pages; none found on 12 Nov 2026 of a page documenting the AI Endpoint client or its operating systems there: none found on 12 Nov 2026 (checked: [knowledge.catonetworks.com](https://knowledge.catonetworks.com/llms.txt)) - A published price for Cato AI Security or AI Endpoint on the Cato product pages; Cato describes only a licensing model: none found on 12 Nov 2026 (checked: [catonetworks.com](https://www.catonetworks.com/platform/cato-ai-endpoint/)) ## Position in each situation | Situation | Position | Total | Leader | | --- | --- | --- | --- | | [Coding agents](/best-for/coding-agents/) | 5th of 16 | 69.0 | Bay 75.5 | | [Regulated data](/best-for/regulated-data/) | 4th of 16 | 70.0 | Microsoft 79.0 | | [Lean team with EDR](/best-for/lean-team-edr/) | 5th of 16 (tied) | 67.5 | Bay 76.5 | | [Staff rollout](/best-for/staff-rollout/) | 5th of 16 | 68.5 | Microsoft 75.5 | | [Board in 30 days](/best-for/board-30-days/) | 4th of 16 (tied) | 68.0 | Bay 77.5 | Aim Security under each situation's own weights. Tied marks an equal total; the order follows the tie-break, the higher score on the heaviest criterion, then name. ## Head-to-heads - [Bay vs Aim Security](/compare/bay-vs-aim-security/): Bay 71.5 to Aim Security 69.0 on the published weights - [Aim Security vs Palo Alto Networks](/compare/aim-security-vs-palo-alto-networks/): Aim Security 69.0 to Palo Alto Networks 45.0 on the published weights **About Bay.** Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). In regulated data, Bay ranks 3rd of 16 at 70.5. In staff rollout, Bay ranks 2nd of 16 at 73.5. Bay is a young company that left stealth in 2026, publishes no price and has little independent testing behind its claims. More: [all 16 platform profiles](/platforms/), [every head-to-head](/compare/), [the six agent surfaces](/surfaces/) and [the six agent actions](/actions/). ## Sources - www.catonetworks.com/platform/cato-ai-endpoint: https://www.catonetworks.com/platform/cato-ai-endpoint/ (read 12 Nov 2026) - www.prnewswire.com/news-releases/cato-networks-launches-cato-ai-endpoint-to-secure-ai-use-beyond-the-browser-302919573.html: https://www.prnewswire.com/news-releases/cato-networks-launches-cato-ai-endpoint-to-secure-ai-use-beyond-the-browser-302919573.html (read 12 Nov 2026) - knowledge.catonetworks.com/docs/product-updates-october-19-2026.md: https://knowledge.catonetworks.com/docs/product-updates-october-19-2026.md (read 13 Nov 2026) - knowledge.catonetworks.com/docs/product-updates-november-9-2026.md: https://knowledge.catonetworks.com/docs/product-updates-november-9-2026.md (read 13 Nov 2026) - knowledge.catonetworks.com/docs/integrating-anthropic-compliance-api-with-cato-ai-security: https://knowledge.catonetworks.com/docs/integrating-anthropic-compliance-api-with-cato-ai-security (read 12 Nov 2026) - techcrunch.com/2024/01/31/aim-security-raises-10m-for-its-genai-security-platform: https://techcrunch.com/2024/01/31/aim-security-raises-10m-for-its-genai-security-platform/ (read 12 Nov 2026) - candid.technology/best-shadow-agent-monitoring-solutions-for-enterprises: https://candid.technology/best-shadow-agent-monitoring-solutions-for-enterprises/ (read 13 Nov 2026) - www.catonetworks.com/platform/agentic-ai-security: https://www.catonetworks.com/platform/agentic-ai-security/ (read 12 Nov 2026) - knowledge.catonetworks.com/docs/product-updates-october-26-2026.md: https://knowledge.catonetworks.com/docs/product-updates-october-26-2026.md (read 13 Nov 2026) - knowledge.catonetworks.com/docs/securing-ai-app-traffic: https://knowledge.catonetworks.com/docs/securing-ai-app-traffic (read 12 Nov 2026) - knowledge.catonetworks.com/docs/product-updates-october-12-2026.md: https://knowledge.catonetworks.com/docs/product-updates-october-12-2026.md (read 13 Nov 2026) - knowledge.catonetworks.com/llms.txt (checked, nothing found): https://knowledge.catonetworks.com/llms.txt (read 12 Nov 2026) - www.catonetworks.com/news/cato-acquires-aim-security-to-extend-sase-leadership-and-secure-enterprise-ai-transformation: https://www.catonetworks.com/news/cato-acquires-aim-security-to-extend-sase-leadership-and-secure-enterprise-ai-transformation/ (read 12 Nov 2026) - www.prnewswire.com/news-releases/cato-networks-achieves-aws-security-competency-in-ai-security-302898714: https://www.prnewswire.com/news-releases/cato-networks-achieves-aws-security-competency-in-ai-security-302898714 (read 12 Nov 2026) - knowledge.catonetworks.com/docs/product-updates-november-2-2026.md: https://knowledge.catonetworks.com/docs/product-updates-november-2-2026.md (read 13 Nov 2026) - www.catonetworks.com/platform/ai-security-for-end-users: https://www.catonetworks.com/platform/ai-security-for-end-users/ (read 12 Nov 2026) - knowledge.catonetworks.com/docs/why-ai-security-policies-may-not-be-enforced-on-platforms-like-microsoft-copilot: https://knowledge.catonetworks.com/docs/why-ai-security-policies-may-not-be-enforced-on-platforms-like-microsoft-copilot (read 12 Nov 2026) - knowledge.catonetworks.com/docs/product-updates-october-5-2026.md: https://knowledge.catonetworks.com/docs/product-updates-october-5-2026.md (read 13 Nov 2026) - knowledge.catonetworks.com/docs/2026-release-notes: https://knowledge.catonetworks.com/docs/2026-release-notes (read 12 Nov 2026) - services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2025-32711: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2025-32711 (read 12 Nov 2026) - thehackernews.com/2025/06/zero-click-ai-vulnerability-exposes.html: https://thehackernews.com/2025/06/zero-click-ai-vulnerability-exposes.html (read 12 Nov 2026) - www.securityweek.com/echoleak-ai-attack-enabled-theft-of-sensitive-data-via-microsoft-365-copilot: https://www.securityweek.com/echoleak-ai-attack-enabled-theft-of-sensitive-data-via-microsoft-365-copilot/ (read 12 Nov 2026) - www.securityweek.com/aim-security-raises-18m-to-secure-customers-implementation-of-ai-apps: https://www.securityweek.com/aim-security-raises-18m-to-secure-customers-implementation-of-ai-apps/ (read 12 Nov 2026) - www.securityweek.com/cato-networks-acquires-ai-security-firm-aim-security: https://www.securityweek.com/cato-networks-acquires-ai-security-firm-aim-security/ (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/platforms/aim-security/ --- # CrowdStrike: 5th of 16 at 66.0, strongest on Sees at 8 of 10 PLATFORM PROFILE, read 13 Nov 2026 **The short answer.** CrowdStrike ranks 5th of 16 AI endpoint security tools at 66.0 of 100. CrowdStrike, Noma Security and Island tie at 66.0; on Sees, the heaviest criterion here, they score 8, 8 and 7, and name order separates equal scores. CrowdStrike stands out on Sees at 8 of 10, and its lowest score is Transparency at 3 of 10. ## Profile - **Owner:** CrowdStrike Holdings, Inc. (NASDAQ: CRWD); acquired Pangea in 2025 - **Product:** Falcon Guardian (CrowdStrike Falcon AI Detection and Response, AIDR) - **Status:** Active under this name, checked 13 Nov 2026 - **Category:** Endpoint platform vendor with AI detection and response (AIDR) for employee AI use and agents - **Deployment:** Existing Falcon endpoint sensor plus agent-level hooks; optional AI gateway and browser - **Operating systems:** Windows, macOS, Linux (agent discovery pre-beta) - **Time to deploy:** none found on 12 Nov 2026 - **Price:** none found on 12 Nov 2026 for Falcon Guardian; endpoint bundles are priced separately (Falcon Enterprise $184.99 per device per year) - **Website:** [crowdstrike.com](https://www.crowdstrike.com/en-us/platform/falcon-guardian-aidr/) Scores out of 10 (weight)Sees 258Stops 257Deployment 156Beyond 106Policy 107Transparency 103Proof 57 ## Strength and weakness **Strength:** Falcon Guardian pairs OS-level sensor telemetry with agent-level hooks that can block actions before they run in Claude Code, Codex, GitHub Copilot, OpenCode and Cursor, plus package blocking in the same sensor. **Weakness:** None found on 13 Nov 2026: a published Falcon Guardian price or public docs; the kill switch acts at the AI gateway while endpoint process termination and endpoint network inspection are pre-beta or coming soon, and the Claude Cowork integration detects without blocking. ## Scores: 66.0 of 100, 5th of 16 | Criterion | Weight | Score | Why this score | | --- | --- | --- | --- | | Sees agentic activity on the endpoint | 25 | 8 | The vendor says it discovers agents, MCP servers, skills, desktop apps, browser extensions and IDEs, and trade press confirms the surfaces. Linux agent discovery is pre-beta. | | Stops the action | 25 | 7 | The vendor says hooks block harmful actions for named coding agents and that its sensor blocks malicious package installs before they run; trade press confirms prompt inspection. Third-party press reads the kill switch as gateway-only for now, and endpoint termination is still coming soon. | | Deployment friction | 15 | 6 | The vendor uses its existing lightweight sensor on Windows, macOS and Linux. No time-to-deploy figure was found on 12 Nov 2026. | | Beyond the endpoint | 10 | 6 | The vendor enforces browser paste and upload policy, trade press confirms discovery of SaaS agents, and the vendor extends to a cloud agent guardrail service. Third-party press reports the cloud workspace integration is a macOS beta focused on visibility and detection without blocking, and SaaS agent coverage is discovery rather than enforcement. | | Policy and governance | 10 | 7 | The vendor offers an agent graph for investigation, a gateway, managed response, audit-trail records of kill switch actions and its own SIEM, which trade press confirms ingests telemetry. An identity-provider integration was none found on 12 Nov 2026. | | Transparency | 10 | 3 | The vendor publishes research on agents and endpoint bundle prices, and trade press says the agent modules are sold through flexible licensing. Docs are login-gated, and a price for the AI product was none found on 12 Nov 2026. | | Proof | 5 | 7 | Third-party press confirms funding for the acquired startup, and the vendor names several customers and an IDC Leader placement. The analyst placement is for endpoint security, and analyst coverage of the AI product was none found on 13 Nov 2026. | CrowdStrike: weighted total 66.0 of 100 on the published weights. ## What does CrowdStrike see on an endpoint? CrowdStrike documents 4 of 6 agent surfaces in full (desktop apps, IDE assistants, command-line agents and MCP servers) and 2 in part (browser extensions and SaaS copilots). Each cell with its note and source: [CrowdStrike on the Agent Surface Map](/agent-surface-map/crowdstrike/). ## Which agent actions can CrowdStrike stop? CrowdStrike documents stopping 2 of 6 agent actions before they complete (data paste and package install) and 4 in part (file read, shell command, network call and credential access). Each cell with its note and source: [CrowdStrike on the Action Control Matrix](/action-control-matrix/crowdstrike/). ## How much does CrowdStrike cost? CrowdStrike publishes no list price. None found on 12 Nov 2026 for Falcon Guardian; endpoint bundles are priced separately (Falcon Enterprise $184.99 per device per year) Price counts toward no score; deployment friction is scored under Deployment. ## Evidence behind the scores - CrowdStrike says Falcon Guardian discovers shadow AI, agents, applications, models, MCP servers, skills and tools on endpoints. (the vendor's own claim; [crowdstrike.com](https://www.crowdstrike.com/en-us/platform/falcon-guardian-aidr/), read 12 Nov 2026) - CrowdStrike says Guardian continuously discovers known and unknown AI agents on supported Windows, macOS and Linux endpoints. (the vendor's own claim; [crowdstrike.com](https://www.crowdstrike.com/en-us/blog/falcon-guardian-defines-next-generation-of-ai-security/), read 12 Nov 2026) - MSSP Alert, covering CrowdStrike's March 2026 RSA announcements, says Shadow AI Discovery for Endpoint surfaces AI apps, agents, LLM runtimes, MCP servers and development tools. (third-party source; [msspalert.com](https://www.msspalert.com/news/crowdstrike-brings-ai-security-to-the-endpoint-and-the-timing-makes-sense), read 12 Nov 2026) - CrowdStrike's Shadow AI Visibility Service blog says its services team finds shadow AI on the endpoint across desktop AI apps, browser extensions, IDEs, packages, MCP servers, models and frameworks. (the vendor's own claim; [crowdstrike.com](https://www.crowdstrike.com/en-us/blog/crowdstrike-shadow-AI-visibility-service/), read 12 Nov 2026) - MSSP Alert reports CrowdStrike sensors detect more than 1,800 distinct AI applications on enterprise devices, nearly 160 million application instances across its customer base. (third-party source; [msspalert.com](https://www.msspalert.com/news/crowdstrike-brings-ai-security-to-the-endpoint-and-the-timing-makes-sense), read 12 Nov 2026) - A November 2026 CrowdStrike post lists Linux agent discovery for Claude Code and Codex as pre-beta, and says Guardian discovers malicious AI skills and MCP servers. (the vendor's own claim; [crowdstrike.com](https://www.crowdstrike.com/en-us/blog/crowdstrike-advances-ai-security-and-safety-falcon-guardian/), read 12 Nov 2026) - CrowdStrike says agent-level hooks for Claude Code, OpenAI Codex, GitHub Copilot, OpenCode and Cursor are available now and can inspect prompts and block malicious activity before harmful actions run. (the vendor's own claim; [crowdstrike.com](https://www.crowdstrike.com/en-us/blog/crowdstrike-advances-ai-security-and-safety-falcon-guardian/), read 12 Nov 2026) - CrowdStrike says Guardian lets organizations define which supported agent types may run on managed endpoints and prevent unauthorized agent types from running. (the vendor's own claim; [crowdstrike.com](https://www.crowdstrike.com/en-us/blog/falcon-guardian-defines-next-generation-of-ai-security/), read 12 Nov 2026) - CrowdStrike says its sensor detects and blocks malicious open-source packages (npm and PyPI) as they reach the endpoint, before embedded code executes, including packages pulled in by AI coding agents. (the vendor's own claim; [crowdstrike.com](https://www.crowdstrike.com/en-us/blog/crowdstrike-extends-endpoint-security-to-stop-supply-chain-attacks/), read 12 Nov 2026) - CrowdStrike says Falcon AIDR blocks credentials, regulated data and other sensitive information before it reaches models, agents or external AI systems. (the vendor's own claim; [crowdstrike.com](https://www.crowdstrike.com/en-us/press-releases/crowdstrike-announces-general-availability-of-falcon-ai-detection-and-response/), read 12 Nov 2026) - MSSP Alert reports AIDR for Endpoint inspects prompts in real time for injection, data leaks and policy violations in desktop AI apps including ChatGPT, Gemini, Claude, DeepSeek, Microsoft Copilot, GitHub Copilot and Cursor. (third-party source; [msspalert.com](https://www.msspalert.com/news/crowdstrike-brings-ai-security-to-the-endpoint-and-the-timing-makes-sense), read 12 Nov 2026) - CrowdStrike lists terminating rogue agent processes on endpoints and preventing restart as coming soon for the kill switch, not yet available. (the vendor's own claim; [crowdstrike.com](https://www.crowdstrike.com/en-us/blog/crowdstrike-introduces-ai-kill-switch/), read 12 Nov 2026) - The kill switch press release says the Falcon sensor stops an agent first at the gateway and soon on the endpoint, so endpoint termination is not yet shipped (limits the endpoint reading of the kill switch). (the vendor's own claim; [crowdstrike.com](https://www.crowdstrike.com/en-us/press-releases/crowdstrike-introduces-ai-kill-switch-for-ai-safety/), read 13 Nov 2026) - 24/7 Wall St. reads the vendor's own wording to mean the kill switch can for now only block an agent's access to models at the gateway, which is the second-source limit on the stops score. (third-party source; [247wallst.com](https://247wallst.com/investing/2026/11/04/crowdstrike-is-spending-495-million-to-police-ai-agents/), read 13 Nov 2026) ## Checks that found nothing - A published time-to-deploy figure for Falcon Guardian on its product page or launch blogs: none found on 12 Nov 2026 (checked: [crowdstrike.com](https://www.crowdstrike.com/en-us/platform/falcon-guardian-aidr/)) - A second (third-party or observed) source documenting Falcon Guardian audit trail, SIEM or identity provider integration, in Google News results for the policy keyword set or on the kill switch coverage read: none found on 13 Nov 2026 (checked: [crowdstrike.com](https://www.crowdstrike.com/en-us/press-releases/crowdstrike-introduces-ai-kill-switch-for-ai-safety/)) - Public product documentation for Falcon Guardian; the documentation link on falcon.crowdstrike.com redirects to a login page: none found on 12 Nov 2026 (checked: [falcon.crowdstrike.com](https://falcon.crowdstrike.com/documentation/)) - A published price for Falcon Guardian or AIDR on the pricing page, which lists only Falcon Go, Pro and Enterprise endpoint bundles: none found on 12 Nov 2026 (checked: [crowdstrike.com](https://www.crowdstrike.com/en-us/pricing/)) - Independent analyst coverage (Gartner, Forrester, IDC, KuppingerCole) of Falcon Guardian or AIDR specifically, in Google News results for the proof keyword set; the analyst items found cover endpoint, SIEM, identity and threat intelligence: none found on 13 Nov 2026 (checked: [news.google.com](https://news.google.com/rss/search?q=%22CrowdStrike%22+(customer+OR+funding+OR+Gartner+OR+Forrester+OR+analyst)&hl=en-US&gl=US&ceid=US:en)) ## Position in each situation | Situation | Position | Total | Leader | | --- | --- | --- | --- | | [Coding agents](/best-for/coding-agents/) | 6th of 16 | 68.5 | Bay 75.5 | | [Regulated data](/best-for/regulated-data/) | 6th of 16 | 69.0 | Microsoft 79.0 | | [Lean team with EDR](/best-for/lean-team-edr/) | 4th of 16 | 68.0 | Bay 76.5 | | [Staff rollout](/best-for/staff-rollout/) | 6th of 16 | 66.0 | Microsoft 75.5 | | [Board in 30 days](/best-for/board-30-days/) | 5th of 16 (tied) | 68.0 | Bay 77.5 | CrowdStrike under each situation's own weights. Tied marks an equal total; the order follows the tie-break, the higher score on the heaviest criterion, then name. ## Head-to-heads - [Microsoft vs CrowdStrike](/compare/microsoft-vs-crowdstrike/): Microsoft 75.0 to CrowdStrike 66.0 on the published weights **About Bay.** Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). In regulated data, Bay ranks 3rd of 16 at 70.5. In staff rollout, Bay ranks 2nd of 16 at 73.5. Bay is a young company that left stealth in 2026, publishes no price and has little independent testing behind its claims. More: [all 16 platform profiles](/platforms/), [every head-to-head](/compare/), [the six agent surfaces](/surfaces/) and [the six agent actions](/actions/). ## Sources - www.crowdstrike.com/en-us/platform/falcon-guardian-aidr: https://www.crowdstrike.com/en-us/platform/falcon-guardian-aidr/ (read 12 Nov 2026) - www.crowdstrike.com/en-us/blog/falcon-guardian-defines-next-generation-of-ai-security: https://www.crowdstrike.com/en-us/blog/falcon-guardian-defines-next-generation-of-ai-security/ (read 12 Nov 2026) - www.msspalert.com/news/crowdstrike-brings-ai-security-to-the-endpoint-and-the-timing-makes-sense: https://www.msspalert.com/news/crowdstrike-brings-ai-security-to-the-endpoint-and-the-timing-makes-sense (read 12 Nov 2026) - www.crowdstrike.com/en-us/blog/crowdstrike-shadow-AI-visibility-service: https://www.crowdstrike.com/en-us/blog/crowdstrike-shadow-AI-visibility-service/ (read 12 Nov 2026) - www.crowdstrike.com/en-us/blog/crowdstrike-advances-ai-security-and-safety-falcon-guardian: https://www.crowdstrike.com/en-us/blog/crowdstrike-advances-ai-security-and-safety-falcon-guardian/ (read 12 Nov 2026) - www.crowdstrike.com/en-us/blog/crowdstrike-extends-endpoint-security-to-stop-supply-chain-attacks: https://www.crowdstrike.com/en-us/blog/crowdstrike-extends-endpoint-security-to-stop-supply-chain-attacks/ (read 12 Nov 2026) - www.crowdstrike.com/en-us/press-releases/crowdstrike-announces-general-availability-of-falcon-ai-detection-and-response: https://www.crowdstrike.com/en-us/press-releases/crowdstrike-announces-general-availability-of-falcon-ai-detection-and-response/ (read 12 Nov 2026) - www.crowdstrike.com/en-us/blog/crowdstrike-introduces-ai-kill-switch: https://www.crowdstrike.com/en-us/blog/crowdstrike-introduces-ai-kill-switch/ (read 12 Nov 2026) - www.crowdstrike.com/en-us/press-releases/crowdstrike-introduces-ai-kill-switch-for-ai-safety: https://www.crowdstrike.com/en-us/press-releases/crowdstrike-introduces-ai-kill-switch-for-ai-safety/ (read 13 Nov 2026) - 247wallst.com/investing/2026/11/04/crowdstrike-is-spending-495-million-to-police-ai-agents: https://247wallst.com/investing/2026/11/04/crowdstrike-is-spending-495-million-to-police-ai-agents/ (read 13 Nov 2026) - www.crowdstrike.com/en-us/blog/crowdstrike-brings-enterprise-browser-security-to-falcon-platform: https://www.crowdstrike.com/en-us/blog/crowdstrike-brings-enterprise-browser-security-to-falcon-platform/ (read 12 Nov 2026) - www.crowdstrike.com/en-us/blog/crowdstrike-secures-ai-agents-inside-claude-cowork: https://www.crowdstrike.com/en-us/blog/crowdstrike-secures-ai-agents-inside-claude-cowork/ (read 13 Nov 2026) - www.crowdstrike.com/en-us/press-releases/crowdstrike-announces-native-ai-security-aws: https://www.crowdstrike.com/en-us/press-releases/crowdstrike-announces-native-ai-security-aws/ (read 13 Nov 2026) - www.crowdstrike.com/en-us/pricing: https://www.crowdstrike.com/en-us/pricing/ (read 12 Nov 2026) - falcon.crowdstrike.com/documentation (checked, nothing found): https://falcon.crowdstrike.com/documentation/ (read 12 Nov 2026) - www.bankinfosecurity.com/crowdstrike-buys-pangea-for-260m-to-guard-enterprise-ai-use: https://www.bankinfosecurity.com/crowdstrike-buys-pangea-for-260m-to-guard-enterprise-ai-use (read 12 Nov 2026) - www.crowdstrike.com/en-us/resources/customer-stories/taxslayer: https://www.crowdstrike.com/en-us/resources/customer-stories/taxslayer/ (read 12 Nov 2026) - www.crowdstrike.com/en-us/resources/customer-stories/so-drama-entertainment: https://www.crowdstrike.com/en-us/resources/customer-stories/so-drama-entertainment/ (read 12 Nov 2026) - news.google.com/rss/search?q=%22CrowdStrike%22+(customer+OR+funding+OR+Gartner+OR+Forrester+OR+analyst)&hl=en-US&gl=US&ceid=US:en (checked, nothing found): https://news.google.com/rss/search?q=%22CrowdStrike%22+(customer+OR+funding+OR+Gartner+OR+Forrester+OR+analyst)&hl=en-US&gl=US&ceid=US:en (read 13 Nov 2026) - www.securityweek.com/island-raises-400-million-at-6-4-billion-valuation: https://www.securityweek.com/island-raises-400-million-at-6-4-billion-valuation/ (read 12 Nov 2026) - securitybrief.co.uk/story/island-named-only-gartner-customers-choice-for-browsers: https://securitybrief.co.uk/story/island-named-only-gartner-customers-choice-for-browsers (read 13 Nov 2026) - github.com/Noma-Security/noma-marketplace: https://github.com/Noma-Security/noma-marketplace (read 12 Nov 2026) - github.com/Noma-Security/noma-marketplace/blob/main/guardrails/README.md: https://github.com/Noma-Security/noma-marketplace/blob/main/guardrails/README.md (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/platforms/crowdstrike/ --- # Noma Security: 6th of 16 at 66.0, strongest on Sees at 8 of 10 PLATFORM PROFILE, read 13 Nov 2026 **The short answer.** Noma Security ranks 6th of 16 AI endpoint security tools at 66.0 of 100. CrowdStrike, Noma Security and Island tie at 66.0; on Sees, the heaviest criterion here, they score 8, 8 and 7, and name order separates equal scores. Noma Security stands out on Sees at 8 of 10, and its lowest score is Transparency at 4 of 10. ## Profile - **Owner:** Noma Security (independent, venture-backed; no parent named on its pages) - **Product:** Noma Agent Security and Governance Platform (AI-SPM, Agent Access Control, AI-DR); Noma for Endpoint AI Agents - **Status:** Active under this name, checked 13 Nov 2026 - **Category:** AI agent security and governance platform with a dedicated endpoint agents module - **Deployment:** Discovery through existing EDR or MDM; runtime enforcement through agent hooks, AI and MCP gateways, SDKs and APIs (vendor-stated) - **Operating systems:** macOS, Linux, Windows - **Time to deploy:** none found on 12 Nov 2026 - **Price:** none found on 12 Nov 2026 (noma.security/pricing returned Not Found; demo-only) - **Website:** [noma.security](https://noma.security) Scores out of 10 (weight)Sees 258Stops 258Deployment 156Beyond 105Policy 105Transparency 104Proof 56 ## Strength and weakness **Strength:** Has a dedicated, current endpoint page: discovery of agents, MCP servers and skills via existing EDR or MDM, plus action-level allow and block policy tied to IdP identity. **Weakness:** Enforcement depends on supported hooks, gateways and APIs per agent, Agent Boundaries are marked coming soon, and no OS list, time to deploy or price was found on 12 Nov 2026. ## Scores: 66.0 of 100, 6th of 16 | Criterion | Weight | Score | Why this score | | --- | --- | --- | --- | | Sees agentic activity on the endpoint | 25 | 8 | The vendor names eight agents and inventories IDEs, MCP servers, skills, hooks and accounts from local files and EDR or MDM data, and an observed public repo hooks named agents. Browser coverage was none found on 12 Nov 2026. | | Stops the action | 25 | 8 | Observed plugin READMEs show block, mask and ask on shell commands, MCP calls, file reads and prompts. The hooks fail open, are in Beta, and some boundaries are marked coming soon. | | Deployment friction | 15 | 6 | Discovery uses existing EDR or MDM, and observed READMEs list macOS, Linux and Windows for the hooks plugin. The vendor says Windows is a one-by-one install today, the observed MDM path covers only macOS and Windows fleets, and a time-to-deploy figure was none found on 13 Nov 2026. | | Beyond the endpoint | 10 | 5 | The vendor sells one control plane across endpoint, SaaS and cloud agents, but documents that its SaaS agent integration for one major platform is monitoring only, after the fact, with no inline blocking. Browser extension or browser-agent control was none found on 13 Nov 2026. | | Policy and governance | 10 | 5 | The vendor gives each agent, MCP server and skill a registry status, limits actions by role, and ties policy to IdP users and groups; discovery uses the existing EDR or MDM. Audit trail detail and a named SIEM integration were none found on 13 Nov 2026. | | Transparency | 10 | 4 | The vendor publishes agent-vulnerability research, with several findings confirmed by third-party press. A published price and public documentation (docs redirect to a login page) were none found on 12 Nov 2026. | | Proof | 5 | 6 | Third-party press confirms a $100 million Series B and paying Fortune 500 customers, and the vendor cites Gartner placements. No named customers were found on 12 Nov 2026. | Noma Security: weighted total 66.0 of 100 on the published weights. ## What does Noma Security see on an endpoint? Noma Security documents 4 of 6 agent surfaces in full (IDE assistants, command-line agents, MCP servers and SaaS copilots) and 1 in part (desktop apps). Nothing was found for browser extensions. Each cell with its note and source: [Noma Security on the Agent Surface Map](/agent-surface-map/noma-security/). ## Which agent actions can Noma Security stop? Noma Security documents stopping 2 of 6 agent actions before they complete (file read and shell command) and 3 in part (network call, credential access and data paste). Nothing was found for package install. Each cell with its note and source: [Noma Security on the Action Control Matrix](/action-control-matrix/noma-security/). ## How much does Noma Security cost? Noma Security publishes no list price. None found on 12 Nov 2026 (noma.security/pricing returned Not Found; demo-only) Price counts toward no score; deployment friction is scored under Deployment. ## Evidence behind the scores - Noma's endpoint page says it finds agents, MCP servers and skills on employee machines; named agents are Claude Cowork, Claude Code and Codex. (the vendor's own claim; [noma.security](https://noma.security/platform/endpoint-agents), read 12 Nov 2026) - Noma's endpoint launch post says it builds a live inventory of agents, IDEs, MCP servers, skills, hooks and connected accounts across managed devices from local configuration files, skills directories, connector history and running processes. (the vendor's own claim; [noma.security](https://noma.security/blog/noma-brings-agent-security-to-every-employee-endpoint), read 12 Nov 2026) - Noma names the coding and productivity agents it covers on endpoints: Claude Code, Claude Cowork, Cursor, Codex, Windsurf, Kiro, Antigravity and OpenClaw. (the vendor's own claim; [noma.security](https://noma.security/blog/noma-brings-agent-security-to-every-employee-endpoint), read 12 Nov 2026) - Noma's MCP security page says it shows installed approved and shadow MCP servers, which agents use them and which tools they expose, and flags supply-chain risks such as unpinned packages. (the vendor's own claim; [noma.security](https://noma.security/solutions/mcp-security), read 12 Nov 2026) - Noma's public marketplace repository on GitHub lists guardrails plugins that hook Claude Code, Codex, Cursor, GitHub Copilot and Kiro, evaluating prompts, shell commands, MCP calls and file access. (observed in public documentation or code; [github.com](https://github.com/Noma-Security/noma-marketplace), read 12 Nov 2026) - Noma says it finds Claude Code by reading local config files for MCP servers and skills, and through CrowdStrike and Microsoft Defender for Endpoint can spot activity where nothing is installed yet. (the vendor's own claim; [noma.security](https://noma.security/blog/securing-claude-ecosystem-how-noma-covers-every-claude-surface), read 13 Nov 2026) - Noma says it finds Claude Cowork (Anthropic desktop environment) through MDM by reading local session files and the scheduled tasks it leaves behind. (the vendor's own claim; [noma.security](https://noma.security/blog/securing-claude-ecosystem-how-noma-covers-every-claude-surface), read 13 Nov 2026) - Noma's Cursor post says it uses Cursor Agent Hooks for inline decisions before actions happen, targeting destructive operations, credential leakage, indirect injection, tool misuse and workspace-trust code execution. (the vendor's own claim; [noma.security](https://noma.security/blog/securing-the-agentic-frontier-noma-unveils-the-first-real-time-agent-runtime-security-for-cursor), read 12 Nov 2026) - Noma's Claude Code plugin README says its hooks act on shell execution, MCP tool execution, file reads and prompt submission, with BLOCK, MASK and ASK policies. (observed in public documentation or code; [github.com](https://github.com/Noma-Security/noma-marketplace/blob/main/guardrails/README.md), read 12 Nov 2026) - Noma's Claude Code plugin README says its file-read control protects.env files and SSH keys from being sent to the model. (observed in public documentation or code; [github.com](https://github.com/Noma-Security/noma-marketplace/blob/main/guardrails/README.md), read 12 Nov 2026) - Noma's Cursor plugin README says ASK verdicts are non-blocking in Cursor while BLOCK verdicts remain enforced. (observed in public documentation or code; [github.com](https://github.com/Noma-Security/noma-marketplace/blob/main/cursor-guardrails/README.md), read 12 Nov 2026) - Noma's marketplace README states a fail-open design: if a hook has no key, no network or bad input, it emits no decision and the agent's default behavior stays in place, so enforcement is not guaranteed on those failures. (observed in public documentation or code; [github.com](https://github.com/Noma-Security/noma-marketplace), read 12 Nov 2026) - Noma's Claude Code plugin README labels Claude Code Hooks as Beta, a status in which Noma may suspend work on the feature. (observed in public documentation or code; [github.com](https://github.com/Noma-Security/noma-marketplace/blob/main/guardrails/README.md), read 12 Nov 2026) - Noma's Agent Boundaries (block, mask, human approval at action time) are labeled 'Coming soon' in its September 28, 2026 endpoint post. (the vendor's own claim; [noma.security](https://noma.security/blog/noma-brings-agent-security-to-every-employee-endpoint), read 12 Nov 2026) ## Checks that found nothing - Operating systems supported for endpoint discovery or enforcement, or a time-to-deploy figure, on Noma's endpoint page, launch post and 90-day rollout guide page: none found on 12 Nov 2026 (checked: [noma.security](https://noma.security/platform/endpoint-agents)) - A Noma page documenting a browser extension or browser-agent control on employee browsers; browser sessions appear only as something agents inherit, plus a CISO guide blog on agentic browsers: none found on 12 Nov 2026 (checked: [noma.security](https://noma.security/blog/noma-brings-agent-security-to-every-employee-endpoint)) - An independent or observed source documenting Noma controls over browsers or SaaS-hosted copilots; Google News query "Noma Security" browser OR SaaS OR copilot returned only vendor releases and unrelated vulnerability news: none found on 13 Nov 2026 (checked: [news.google.com](https://news.google.com/rss/search?q=%22Noma+Security%22+browser+OR+SaaS+OR+copilot)) - An independent or observed source documenting Noma SIEM, Splunk or Okta integrations; Google News query "Noma Security" SIEM OR Splunk OR Okta OR audit returned hiring news and list articles only: none found on 13 Nov 2026 (checked: [news.google.com](https://news.google.com/rss/search?q=%22Noma+Security%22+SIEM+OR+Splunk+OR+Okta+OR+audit)) - A published price (noma.security/pricing returned Not Found) or a public documentation site; the site offers demos, blog, Noma Labs and white papers: none found on 12 Nov 2026 (checked: [noma.security](https://noma.security/pricing)) ## Position in each situation | Situation | Position | Total | Leader | | --- | --- | --- | --- | | [Coding agents](/best-for/coding-agents/) | 4th of 16 | 69.5 | Bay 75.5 | | [Regulated data](/best-for/regulated-data/) | 9th of 16 (tied) | 64.0 | Microsoft 79.0 | | [Lean team with EDR](/best-for/lean-team-edr/) | 6th of 16 (tied) | 67.5 | Bay 76.5 | | [Staff rollout](/best-for/staff-rollout/) | 9th of 16 (tied) | 63.5 | Microsoft 75.5 | | [Board in 30 days](/best-for/board-30-days/) | 6th of 16 (tied) | 66.5 | Bay 77.5 | Noma Security under each situation's own weights. Tied marks an equal total; the order follows the tie-break, the higher score on the heaviest criterion, then name. **About Bay.** Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). In regulated data, Bay ranks 3rd of 16 at 70.5. In staff rollout, Bay ranks 2nd of 16 at 73.5. Bay is a young company that left stealth in 2026, publishes no price and has little independent testing behind its claims. More: [all 16 platform profiles](/platforms/), [every head-to-head](/compare/), [the six agent surfaces](/surfaces/) and [the six agent actions](/actions/). ## Sources - noma.security/platform/endpoint-agents: https://noma.security/platform/endpoint-agents (read 12 Nov 2026) - noma.security/blog/noma-brings-agent-security-to-every-employee-endpoint: https://noma.security/blog/noma-brings-agent-security-to-every-employee-endpoint (read 12 Nov 2026) - noma.security/solutions/mcp-security: https://noma.security/solutions/mcp-security (read 12 Nov 2026) - github.com/Noma-Security/noma-marketplace: https://github.com/Noma-Security/noma-marketplace (read 12 Nov 2026) - noma.security/blog/securing-claude-ecosystem-how-noma-covers-every-claude-surface: https://noma.security/blog/securing-claude-ecosystem-how-noma-covers-every-claude-surface (read 13 Nov 2026) - noma.security/blog/securing-the-agentic-frontier-noma-unveils-the-first-real-time-agent-runtime-security-for-cursor: https://noma.security/blog/securing-the-agentic-frontier-noma-unveils-the-first-real-time-agent-runtime-security-for-cursor (read 12 Nov 2026) - github.com/Noma-Security/noma-marketplace/blob/main/guardrails/README.md: https://github.com/Noma-Security/noma-marketplace/blob/main/guardrails/README.md (read 12 Nov 2026) - github.com/Noma-Security/noma-marketplace/blob/main/cursor-guardrails/README.md: https://github.com/Noma-Security/noma-marketplace/blob/main/cursor-guardrails/README.md (read 12 Nov 2026) - noma.security: https://noma.security/ (read 12 Nov 2026) - noma.security/blog/noma-grok-bot: https://noma.security/blog/noma-grok-bot (read 12 Nov 2026) - noma.security/integrations/claude-enterprise: https://noma.security/integrations/claude-enterprise (read 13 Nov 2026) - news.google.com/rss/search?q=%22Noma+Security%22+browser+OR+SaaS+OR+copilot (checked, nothing found): https://news.google.com/rss/search?q=%22Noma+Security%22+browser+OR+SaaS+OR+copilot (read 13 Nov 2026) - finance.yahoo.com/technology/ai/articles/noma-extends-agent-security-governance-100000047.html: https://finance.yahoo.com/technology/ai/articles/noma-extends-agent-security-governance-100000047.html (read 12 Nov 2026) - news.google.com/rss/search?q=%22Noma+Security%22+SIEM+OR+Splunk+OR+Okta+OR+audit (checked, nothing found): https://news.google.com/rss/search?q=%22Noma+Security%22+SIEM+OR+Splunk+OR+Okta+OR+audit (read 13 Nov 2026) - finance.yahoo.com/technology/ai/articles/noma-labs-uncovers-critical-security-130000444.html: https://finance.yahoo.com/technology/ai/articles/noma-labs-uncovers-critical-security-130000444.html (read 12 Nov 2026) - thehackernews.com/2025/06/langchain-langsmith-bug-let-hackers.html: https://thehackernews.com/2025/06/langchain-langsmith-bug-let-hackers.html (read 12 Nov 2026) - cybernews.com/security/claude-code-preview-pane-secret-leak-flaw: https://cybernews.com/security/claude-code-preview-pane-secret-leak-flaw/ (read 13 Nov 2026) - www.helpnetsecurity.com/2026/10/29/research-ai-assistant-browser-vulnerabilities: https://www.helpnetsecurity.com/2026/10/29/research-ai-assistant-browser-vulnerabilities/ (read 13 Nov 2026) - noma.security/pricing (checked, nothing found): https://noma.security/pricing (read 12 Nov 2026) - docs.noma.security: https://docs.noma.security (read 12 Nov 2026) - pulse2.com/noma-security-100-million-raised-for-ai-agent-security-platform: https://pulse2.com/noma-security-100-million-raised-for-ai-agent-security-platform/ (read 12 Nov 2026) - techcrunch.com/2024/10/31/noma-is-building-tools-to-spot-security-issues-with-ai-apps: https://techcrunch.com/2024/10/31/noma-is-building-tools-to-spot-security-issues-with-ai-apps/ (read 12 Nov 2026) - noma.security/blog/noma-security-raises-100m-to-drive-adoption-of-ai-agent-security: https://noma.security/blog/noma-security-raises-100m-to-drive-adoption-of-ai-agent-security (read 12 Nov 2026) - finance.yahoo.com/technology/ai/articles/noma-named-market-shaper-gartner-050000887.html: https://finance.yahoo.com/technology/ai/articles/noma-named-market-shaper-gartner-050000887.html (read 12 Nov 2026) - www.msspalert.com/news/crowdstrike-brings-ai-security-to-the-endpoint-and-the-timing-makes-sense: https://www.msspalert.com/news/crowdstrike-brings-ai-security-to-the-endpoint-and-the-timing-makes-sense (read 12 Nov 2026) - www.securityweek.com/island-raises-400-million-at-6-4-billion-valuation: https://www.securityweek.com/island-raises-400-million-at-6-4-billion-valuation/ (read 12 Nov 2026) - securitybrief.co.uk/story/island-named-only-gartner-customers-choice-for-browsers: https://securitybrief.co.uk/story/island-named-only-gartner-customers-choice-for-browsers (read 13 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/platforms/noma-security/ --- # Island: 7th of 16 at 66.0, strongest on Beyond at 8 of 10 PLATFORM PROFILE, read 13 Nov 2026 **The short answer.** Island ranks 7th of 16 AI endpoint security tools at 66.0 of 100. CrowdStrike, Noma Security and Island tie at 66.0; on Sees, the heaviest criterion here, they score 8, 8 and 7, and name order separates equal scores. Island stands out on Beyond at 8 of 10, and its lowest score is Transparency at 3 of 10. ## Profile - **Owner:** Island Technology, Inc., independent and privately held; no parent or acquirer named on its pages - **Product:** Island Enterprise Browser, Island Extension, Island Desktop and Enterprise AI (AI governance, AI runtime protection, MCP gateway) - **Status:** Active under this name, checked 13 Nov 2026 - **Category:** Enterprise browser and agentic control plane spanning browser, extension, endpoint agent, network and MCP gateway - **Deployment:** Enterprise browser, browser extension, endpoint agent (Windows service, macOS daemon) pushed via MDM, network service, MCP gateway and API integrations, as published - **Operating systems:** Windows, macOS - **Time to deploy:** No vendor-published figure found on 12 Nov 2026; one customer quote says a full rollout took a couple of weeks (the vendor's own claim) - **Price:** none found on 12 Nov 2026 (island.io/pricing returns 404; site offers Request a Quote) - **Website:** [island.io](https://www.island.io) Scores out of 10 (weight)Sees 257Stops 257Deployment 156Beyond 108Policy 107Transparency 103Proof 58 ## Strength and weakness **Strength:** Island combines an endpoint agent for Windows and macOS with a browser, extension and MCP gateway, and documents blocking of Bash execution, file reads and network connections as well as clipboard DLP. **Weakness:** None found on 12 Nov 2026: a public price, public documentation site or release notes; how the endpoint agent intercepts CLI agents other than Claude Code is not documented in detail. ## Scores: 66.0 of 100, 7th of 16 | Criterion | Weight | Score | Why this score | | --- | --- | --- | --- | | Sees agentic activity on the endpoint | 25 | 7 | The vendor describes an endpoint inventory of agents, MCP servers, skills, packages, browser and IDE extensions and hooks, with named coding agents. Third-party coverage names the product lines for desktop apps and local agentic resources, but a second source confirming the inventory detail was none found on 13 Nov 2026. | | Stops the action | 25 | 7 | The vendor names blocking of Bash execution, file reads and outbound connections before an agent acts, plus gateway mitigation of injection. Third-party coverage confirms only human paste and download controls, and an independent test of agent-action blocking was none found on 13 Nov 2026. | | Deployment friction | 15 | 6 | The vendor pushes a Windows service and macOS daemon via MDM, and a customer reports a rollout of a couple of weeks. A Linux build, a vendor-published deploy time and an independent source on deployment were none found on 13 Nov 2026. | | Beyond the endpoint | 10 | 8 | The vendor says the same policy reaches consumer browsers, agent browser sessions in the cloud, MCP calls and a SaaS agent platform, and third-party press reports policy enforced when endpoint or cloud agents call the browser, with data, credential and output controls. The third-party piece restates the vendor's launch rather than testing it. | | Policy and governance | 10 | 7 | The vendor documents one policy framework for humans and agents, audit logs that tell them apart, gateway logs exportable to SIEM, and EDR and MDM integrations. All of it is vendor claim; an independent source on the integrations was none found on 13 Nov 2026. | | Transparency | 10 | 3 | The vendor publishes agent supply-chain and lookalike-domain research. A public pricing page and a public documentation site were none found on 12 Nov 2026. | | Proof | 5 | 8 | Third-party press confirms more than $1 billion raised and about 450 customers, and the vendor names customers such as Pfizer and TaskUs. Third-party press also confirms the Gartner mention rests on peer reviews, not an analyst ranking. | Island: weighted total 66.0 of 100 on the published weights. ## What does Island see on an endpoint? Island documents 4 of 6 agent surfaces in full (desktop apps, IDE assistants, MCP servers and browser extensions) and 2 in part (command-line agents and SaaS copilots). Each cell with its note and source: [Island on the Agent Surface Map](/agent-surface-map/island/). ## Which agent actions can Island stop? Island documents stopping 4 of 6 agent actions before they complete (file read, shell command, network call and data paste) and 2 in part (credential access and package install). Each cell with its note and source: [Island on the Action Control Matrix](/action-control-matrix/island/). ## How much does Island cost? Island publishes no list price. None found on 12 Nov 2026 (island.io/pricing returns 404; site offers Request a Quote) Price counts toward no score; deployment friction is scored under Deployment. ## Evidence behind the scores - Island says its AI product discovers AI tools, non-human identities, agents, MCP servers and Bash commands, and scans installed browser extensions and code packages. (the vendor's own claim; [island.io](https://www.island.io/product/ai), read 12 Nov 2026) - Island says its Agentic Endpoint Posture builds an inventory of agents on the endpoint including MCP servers, skills, code packages, browser extensions, IDE extensions and hooks. (the vendor's own claim; [island.io](https://www.island.io/blog/expanding-island-to-govern-your-ai-workforce), read 12 Nov 2026) - Island says it governs browser AI tools, coding agents such as Cursor, Windsurf and Cline, desktop agents, autonomous agents, MCP servers and Bash commands. (the vendor's own claim; [island.io](https://www.island.io/product/ai), read 12 Nov 2026) - Island says Claude Code, Claude Cowork and Claude for Microsoft 365 sessions are read through the Anthropic Compliance API, with prompts, tool calls, skills and artifacts tied to a verified user. (the vendor's own claim; [island.io](https://www.island.io/blog/claude-code-cowork-and-claude-for-microsoft-365-now-report-into-island), read 12 Nov 2026) - Island says Agentic Endpoint Posture gives a live inventory of every AI asset on every device, with a risk score and evidence for each. (the vendor's own claim; [island.io](https://www.island.io/blog/see-score-and-control-every-ai-asset-on-your-endpoints), read 12 Nov 2026) - SecurityWeek describes Island's platform as an agentic control plane meant to govern both human interactions and automated workflows across endpoints, software and networks. (third-party source; [securityweek.com](https://www.securityweek.com/island-raises-400-million-at-6-4-billion-valuation/), read 12 Nov 2026) - SecurityBrief UK (October 28, 2026) describes, from company information, an Island Extension for consumer browsers, Island Desktop for desktop applications and local agentic resources, and Island Network. (third-party source; [securitybrief.co.uk](https://securitybrief.co.uk/story/island-named-only-gartner-customers-choice-for-browsers), read 13 Nov 2026) - Island says AI Runtime Protection can respond to risky agent actions by allowing with logging, prompting the user for confirmation, or blocking the action. (the vendor's own claim; [island.io](https://www.island.io/blog/ai-runtime-protection-governing-what-agents-do-not-just-what-theyre-told), read 12 Nov 2026) - Island lists blocking of prompt injections, malicious URLs and unauthorized Bash execution before an agent acts, as vendor claims. (the vendor's own claim; [island.io](https://www.island.io/product/ai), read 12 Nov 2026) - Island describes a chain where a flagged skill is never run, and otherwise the file read on the endpoint or the outbound connection on the network can still be blocked. (the vendor's own claim; [island.io](https://www.island.io/blog/claude-code-cowork-and-claude-for-microsoft-365-now-report-into-island), read 12 Nov 2026) - Island says per-tool policy can limit agents by action type, and grant time-boxed access instead of inherited standing permissions. (the vendor's own claim; [island.io](https://www.island.io/product/ai), read 12 Nov 2026) - Island says its Claude Inference Hooks integration lets prompts and tool results on claude.ai, Cowork and Claude Code be checked by Island's policy engine before the model runs. (the vendor's own claim; [island.io](https://www.island.io/blog/island-integrates-with-claudes-new-inference-hooks-for-inline-ai-security), read 12 Nov 2026) - SecurityWeek reports Island's DLP rules can block unapproved clipboard transfers, downloads and screen captures where work occurs (human data paste and download controls, not agent actions specifically). (third-party source; [securityweek.com](https://www.securityweek.com/island-raises-400-million-at-6-4-billion-valuation/), read 12 Nov 2026) - Island says its LLM Gateway can detect and mitigate AI-related threats such as prompt injection and context poisoning on model calls (vendor claim, blog dated November 5, 2026). (the vendor's own claim; [island.io](https://www.island.io/blog/island-llm-gateway-one-control-point-for-every-model-call), read 13 Nov 2026) ## Checks that found nothing - An independent test or third-party report confirming Island blocks agent Bash, file read or network actions on the endpoint; Google News query C2 returned only funding, award and older items: none found on 13 Nov 2026 (checked: [news.google.com](https://news.google.com/rss/search?q=%22Island%22+enterprise+browser+(block+OR+runtime+OR+enforcement)&hl=en-US&gl=US&ceid=US:en)) - An independent (non-Island) source documenting Island Desktop operating system coverage or time to deploy for the AI features; Google News query C3 returned only older browser coverage and vendor items: none found on 13 Nov 2026 (checked: [news.google.com](https://news.google.com/rss/search?q=%22Island%22+enterprise+browser+(agentless+OR+deploy+OR+macOS+OR+Windows)&hl=en-US&gl=US&ceid=US:en)) - An independent (non-Island) source documenting Island's SIEM, Splunk or Okta integrations or audit trail for agents; Google News query C5 returned funding and 2023 browser items only: none found on 13 Nov 2026 (checked: [news.google.com](https://news.google.com/rss/search?q=%22Island%22+enterprise+browser+(SIEM+OR+Splunk+OR+Okta+OR+audit)&hl=en-US&gl=US&ceid=US:en)) - A public pricing page; island.io/pricing returns 404 and the site offers Request a Quote instead: none found on 12 Nov 2026 (checked: [island.io](https://www.island.io/pricing)) - A public documentation site; docs.island.io returned 404 and release notes at island.io/release-notes returned 404: none found on 12 Nov 2026 (checked: [docs.island.io](https://docs.island.io/)) ## Position in each situation | Situation | Position | Total | Leader | | --- | --- | --- | --- | | [Coding agents](/best-for/coding-agents/) | 7th of 16 | 67.5 | Bay 75.5 | | [Regulated data](/best-for/regulated-data/) | 5th of 16 | 69.5 | Microsoft 79.0 | | [Lean team with EDR](/best-for/lean-team-edr/) | 7th of 16 | 66.5 | Bay 76.5 | | [Staff rollout](/best-for/staff-rollout/) | 4th of 16 | 69.5 | Microsoft 75.5 | | [Board in 30 days](/best-for/board-30-days/) | 7th of 16 (tied) | 66.5 | Bay 77.5 | Island under each situation's own weights. Tied marks an equal total; the order follows the tie-break, the higher score on the heaviest criterion, then name. ## Head-to-heads - [Bay vs Island](/compare/bay-vs-island/): Bay 71.5 to Island 66.0 on the published weights - [Island vs LayerX](/compare/island-vs-layerx/): Island 66.0 to LayerX 50.0 on the published weights - [Island vs Palo Alto Networks](/compare/island-vs-palo-alto-networks/): Island 66.0 to Palo Alto Networks 45.0 on the published weights **About Bay.** Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). In regulated data, Bay ranks 3rd of 16 at 70.5. In staff rollout, Bay ranks 2nd of 16 at 73.5. Bay is a young company that left stealth in 2026, publishes no price and has little independent testing behind its claims. More: [all 16 platform profiles](/platforms/), [every head-to-head](/compare/), [the six agent surfaces](/surfaces/) and [the six agent actions](/actions/). ## Sources - www.island.io/product/ai: https://www.island.io/product/ai (read 12 Nov 2026) - www.island.io/blog/expanding-island-to-govern-your-ai-workforce: https://www.island.io/blog/expanding-island-to-govern-your-ai-workforce (read 12 Nov 2026) - www.island.io/blog/claude-code-cowork-and-claude-for-microsoft-365-now-report-into-island: https://www.island.io/blog/claude-code-cowork-and-claude-for-microsoft-365-now-report-into-island (read 12 Nov 2026) - www.island.io/blog/see-score-and-control-every-ai-asset-on-your-endpoints: https://www.island.io/blog/see-score-and-control-every-ai-asset-on-your-endpoints (read 12 Nov 2026) - www.securityweek.com/island-raises-400-million-at-6-4-billion-valuation: https://www.securityweek.com/island-raises-400-million-at-6-4-billion-valuation/ (read 12 Nov 2026) - securitybrief.co.uk/story/island-named-only-gartner-customers-choice-for-browsers: https://securitybrief.co.uk/story/island-named-only-gartner-customers-choice-for-browsers (read 13 Nov 2026) - www.island.io/blog/ai-runtime-protection-governing-what-agents-do-not-just-what-theyre-told: https://www.island.io/blog/ai-runtime-protection-governing-what-agents-do-not-just-what-theyre-told (read 12 Nov 2026) - www.island.io/blog/island-integrates-with-claudes-new-inference-hooks-for-inline-ai-security: https://www.island.io/blog/island-integrates-with-claudes-new-inference-hooks-for-inline-ai-security (read 12 Nov 2026) - www.island.io/blog/island-llm-gateway-one-control-point-for-every-model-call: https://www.island.io/blog/island-llm-gateway-one-control-point-for-every-model-call (read 13 Nov 2026) - news.google.com/rss/search?q=%22Island%22+enterprise+browser+(block+OR+runtime+OR+enforcement)&hl=en-US&gl=US&ceid=US:en (checked, nothing found): https://news.google.com/rss/search?q=%22Island%22+enterprise+browser+(block+OR+runtime+OR+enforcement)&hl=en-US&gl=US&ceid=US:en (read 13 Nov 2026) - www.island.io/enterprise-desktop: https://www.island.io/enterprise-desktop (read 12 Nov 2026) - www.island.io/about: https://www.island.io/about (read 12 Nov 2026) - news.google.com/rss/search?q=%22Island%22+enterprise+browser+(agentless+OR+deploy+OR+macOS+OR+Windows)&hl=en-US&gl=US&ceid=US:en (checked, nothing found): https://news.google.com/rss/search?q=%22Island%22+enterprise+browser+(agentless+OR+deploy+OR+macOS+OR+Windows)&hl=en-US&gl=US&ceid=US:en (read 13 Nov 2026) - www.island.io/product/enterprise-browser-extension: https://www.island.io/product/enterprise-browser-extension (read 12 Nov 2026) - www.island.io/press/island-brings-the-enterprise-browser-to-ai-agents: https://www.island.io/press/island-brings-the-enterprise-browser-to-ai-agents (read 12 Nov 2026) - www.island.io/blog/island-integrates-with-the-claude-compliance-api: https://www.island.io/blog/island-integrates-with-the-claude-compliance-api (read 12 Nov 2026) - securitybrief.co.uk/story/island-launches-enterprise-browser-to-control-ai-agents: https://securitybrief.co.uk/story/island-launches-enterprise-browser-to-control-ai-agents (read 13 Nov 2026) - www.island.io/blog/whats-new-in-island-ai-threat-intelligence: https://www.island.io/blog/whats-new-in-island-ai-threat-intelligence (read 13 Nov 2026) - news.google.com/rss/search?q=%22Island%22+enterprise+browser+(SIEM+OR+Splunk+OR+Okta+OR+audit)&hl=en-US&gl=US&ceid=US:en (checked, nothing found): https://news.google.com/rss/search?q=%22Island%22+enterprise+browser+(SIEM+OR+Splunk+OR+Okta+OR+audit)&hl=en-US&gl=US&ceid=US:en (read 13 Nov 2026) - www.island.io/blog/agentbaiting-how-800-fake-ai-skills-and-mcp-servers-delivered-malware: https://www.island.io/blog/agentbaiting-how-800-fake-ai-skills-and-mcp-servers-delivered-malware (read 12 Nov 2026) - www.island.io/blog/typoprompting-from-ai-typos-to-malware: https://www.island.io/blog/typoprompting-from-ai-typos-to-malware (read 13 Nov 2026) - www.island.io/pricing (checked, nothing found): https://www.island.io/pricing (read 12 Nov 2026) - docs.island.io (checked, nothing found): https://docs.island.io/ (read 12 Nov 2026) - www.securityweek.com/island-banks-250m-series-e-for-enterprise-browser: https://www.securityweek.com/island-banks-250m-series-e-for-enterprise-browser/ (read 12 Nov 2026) - www.island.io/blog/taskus-runs-on-default-deny-it-still-said-yes-to-ai: https://www.island.io/blog/taskus-runs-on-default-deny-it-still-said-yes-to-ai (read 12 Nov 2026) - www.island.io/press: https://www.island.io/press (read 12 Nov 2026) - betanews.com/2025/12/17/governing-ai-where-work-actually-happens-qa: https://betanews.com/2025/12/17/governing-ai-where-work-actually-happens-qa/ (read 12 Nov 2026) - www.msspalert.com/news/crowdstrike-brings-ai-security-to-the-endpoint-and-the-timing-makes-sense: https://www.msspalert.com/news/crowdstrike-brings-ai-security-to-the-endpoint-and-the-timing-makes-sense (read 12 Nov 2026) - github.com/Noma-Security/noma-marketplace: https://github.com/Noma-Security/noma-marketplace (read 12 Nov 2026) - github.com/Noma-Security/noma-marketplace/blob/main/guardrails/README.md: https://github.com/Noma-Security/noma-marketplace/blob/main/guardrails/README.md (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/platforms/island/ --- # Zenity: 8th of 16 at 65.0, strongest on Transparency at 7 of 10 PLATFORM PROFILE, read 13 Nov 2026 **The short answer.** Zenity ranks 8th of 16 AI endpoint security tools at 65.0 of 100. Zenity stands out on Transparency at 7 of 10, and its lowest score is Deployment at 5 of 10. Zenity shows 4 of 6 agent surfaces in full and can stop 3 of 6 agent actions, as documented. ## Profile - **Owner:** Zenity Ltd. (independent, venture-backed; no parent named on its pages) - **Product:** Zenity AI Agent Security and Governance Platform (Surface, Enforce, Protect), incl. Personal and Coding Agents - **Status:** Active under this name, checked 13 Nov 2026 - **Category:** AI agent security and governance platform; strongest on SaaS and cloud-built agents, with a hook-based coding-agent module - **Deployment:** Agentless for coding agents via OpenTelemetry and native agent hooks plus an MCP gateway (vendor blog); SaaS and cloud agents connect through API integrations - **Operating systems:** none found on 12 Nov 2026 - **Time to deploy:** none found on 12 Nov 2026 - **Price:** none found on 12 Nov 2026 (zenity.io/pricing returned 404; demo-only) - **Website:** [zenity.io](https://zenity.io) Scores out of 10 (weight)Sees 256Stops 257Deployment 155Beyond 107Policy 107Transparency 107Proof 58 ## Strength and weakness **Strength:** Documents pre-execution blocking of coding-agent tool calls through native hooks and an MCP gateway, including credential-read-then-network sequences. **Weakness:** Endpoint coverage is described as hook and OpenTelemetry based with no OS list, no time to deploy and no price found on 12 Nov 2026, and most of its breadth is SaaS and cloud agents. ## Scores: 65.0 of 100, 8th of 16 | Criterion | Weight | Score | Why this score | | --- | --- | --- | --- | | Sees agentic activity on the endpoint | 25 | 6 | The vendor names coding agents, inventories MCP connections and permission settings, and sees skills and plugins through a SaaS API. Browser extension or desktop app inventory, and an independent source for endpoint discovery, were none found on 13 Nov 2026. | | Stops the action | 25 | 7 | The vendor says hooks block tool calls before they run, naming sensitive file reads, network actions, secret exfiltration and malicious package installs. No independent test or public documentation confirming this was found on 12 Nov 2026. | | Deployment friction | 15 | 5 | The vendor says coding-agent coverage is agentless through hooks and OpenTelemetry. An operating-system list and a time-to-deploy figure were none found on 13 Nov 2026. | | Beyond the endpoint | 10 | 7 | The vendor says one policy spans SaaS agents, cloud-built agents and coding agents, and adds inline control of a shared agent in Slack. Browser coverage and an independent confirmation of the shared policy were none found on 13 Nov 2026. | | Policy and governance | 10 | 7 | The vendor lists Okta, EDR and SIEM integrations, SSO, deterministic policy and audit trails, and an alliance membership. All of it is vendor-stated, and an independent or public-docs confirmation was none found on 13 Nov 2026. | | Transparency | 10 | 7 | The vendor runs a documentation site and publishes agent-vulnerability research that third-party press covers, including a new notice on a cloud agent platform. A published price was none found on 12 Nov 2026. | | Proof | 5 | 8 | Third-party press confirms a $125 million round, the vendor names customers such as SoftBank on the record, and it cites Gartner reports including a Market Shaper placement. The customer names and analyst citations are vendor-published. | Zenity: weighted total 65.0 of 100 on the published weights. ## What does Zenity see on an endpoint? Zenity documents 4 of 6 agent surfaces in full (IDE assistants, command-line agents, MCP servers and SaaS copilots) and 2 in part (desktop apps and browser extensions). Each cell with its note and source: [Zenity on the Agent Surface Map](/agent-surface-map/zenity/). ## Which agent actions can Zenity stop? Zenity documents stopping 3 of 6 agent actions before they complete (shell command, network call and credential access) and 3 in part (file read, data paste and package install). Each cell with its note and source: [Zenity on the Action Control Matrix](/action-control-matrix/zenity/). ## How much does Zenity cost? Zenity publishes no list price. None found on 12 Nov 2026 (zenity.io/pricing returned 404; demo-only) Price counts toward no score; deployment friction is scored under Deployment. ## Evidence behind the scores - Zenity says its Personal and Coding Agents coverage discovers coding and personal agents across developer laptops and desktops plus the MCP servers and tools they connect to; named agents are Claude Code and Cowork, ChatGPT Codex, GitHub Copilot and Cursor. (the vendor's own claim; [zenity.io](https://zenity.io/use-cases/agent-type/coding-personal-agents), read 12 Nov 2026) - A Zenity blog says its posture management monitors permission-mode settings (such as Claude Code bypassPermissions, Codex 'never', Cursor 'Run Everything') and sandbox settings across coding agents. (the vendor's own claim; [zenity.io](https://zenity.io/blog/coding-agent-attack-surface), read 12 Nov 2026) - Zenity says it inventories MCP server connections across the developer estate and flags unreviewed connections and the permission scope each server inherits. (the vendor's own claim; [zenity.io](https://zenity.io/blog/coding-agent-attack-surface), read 12 Nov 2026) - Zenity says its Claude Enterprise integration (via the Compliance API) lets security teams discover and govern MCP servers, plugins and skills across Claude Code, Cowork and Chat. (the vendor's own claim; [zenity.io](https://zenity.io/press-release/zenity-extends-ai-agent-security-and-governance-to-claude-enterprise), read 12 Nov 2026) - Zenity says that through native agent hooks and its MCP gateway it can block or modify a dangerous tool call before it executes, for coding agents including Claude Code, Codex, Copilot and Cursor. (the vendor's own claim; [zenity.io](https://zenity.io/use-cases/agent-type/coding-personal-agents), read 12 Nov 2026) - A Zenity blog says it detects and blocks a session where an agent reads a sensitive file (.env, SSH key, credential store) and then attempts a new network action, at the PreToolUse hook before the call executes. (the vendor's own claim; [zenity.io](https://zenity.io/blog/coding-agent-attack-surface), read 12 Nov 2026) - Zenity says it prevents risks such as secret exfiltration, malicious package installs and unsafe tool chaining before they lead to compromise. (the vendor's own claim; [zenity.io](https://zenity.io/use-cases/agent-type/coding-personal-agents), read 12 Nov 2026) - Zenity's MCP page says it can allow, modify or block a tool call based on live context. (the vendor's own claim; [zenity.io](https://zenity.io/platform/mcp-security), read 12 Nov 2026) - A Zenity blog says coding-agent coverage is agentless, via OpenTelemetry and native hooks, with no endpoint sensor required. (the vendor's own claim; [zenity.io](https://zenity.io/blog/coding-agent-attack-surface), read 12 Nov 2026) - Zenity says MCP governance deploys by publishing a server and registering a single gateway URL, with agents continuing unchanged. (the vendor's own claim; [zenity.io](https://zenity.io/platform/mcp-security), read 12 Nov 2026) - Zenity says its platform spans agentic SaaS (Agentforce, Copilot Studio), cloud and homegrown agents (AWS Bedrock, Google Vertex AI) and personal and coding agents, with unified policy. (the vendor's own claim; [zenity.io](https://zenity.io/), read 12 Nov 2026) - Zenity's Series C release says its security layer covers Microsoft Copilot, ChatGPT Enterprise and Gemini, coding agents like Claude, Codex and Cursor, and custom agents on Bedrock, AgentCore, Foundry and Vertex AI. Most of this breadth is SaaS and cloud-built agents rather than endpoints. (the vendor's own claim; [zenity.io](https://zenity.io/press-release/zenity-raises-125-million-to-secure-the-era-of-1-billion-ai-agents), read 12 Nov 2026) - Zenity announced on June 12, 2026 an integration with Claude's Compliance API that extends its governance to Claude Enterprise, bringing a vendor-hosted SaaS agent platform into the same policy plane. (the vendor's own claim; [zenity.io](https://zenity.io/press-release/zenity-extends-ai-agent-security-and-governance-to-claude-enterprise), read 12 Nov 2026) - Zenity's October 8, 2026 blog says its Claude Tag plugin builds an inventory of the shared Claude agent in Slack and scopes it per channel to reviewed skills, MCP servers and connectors, stopping mutating or irreversible tool calls inline; this is a SaaS-hosted agent, not an endpoint. (the vendor's own claim; [zenity.io](https://zenity.io/blog/zenity-for-claude-tag-securing-the-shared-agent-in-your-slack-workspace), read 13 Nov 2026) ## Checks that found nothing - An independent (third-party or observed) source confirming Zenity's discovery of coding agents, MCP servers or skills on employee endpoints; searches of Google News, Bing News, the blog index and the docs navigation returned only vendor pages, and docs pages for coding-agent integrations redirect to a login: none found on 13 Nov 2026 (checked: [zenity.io](https://zenity.io/blog)) - Independent trade-press testing or a public documentation page confirming Zenity's endpoint blocking of coding-agent actions; docs.zenity.io user-guide pages (coding agents, inline prevention, malicious package protection) redirect to a customer login: none found on 12 Nov 2026 (checked: [docs.zenity.io](https://docs.zenity.io/user-guide/coding-agents)) - Operating systems supported for coding-agent coverage, or a time-to-deploy figure, on Zenity's coding-agents page and blog: none found on 12 Nov 2026 (checked: [zenity.io](https://zenity.io/use-cases/agent-type/coding-personal-agents)) - An independent source or public docs page stating operating systems or time to deploy for Zenity's endpoint and coding-agent coverage; docs.zenity.io coding-agent pages redirect to a login: none found on 13 Nov 2026 (checked: [docs.zenity.io](https://docs.zenity.io/)) - An independent test or write-up confirming that one Zenity policy spans endpoint coding agents, browsers and SaaS agents; news results cover funding and research, not cross-surface policy: none found on 13 Nov 2026 (checked: [docs.zenity.io](https://docs.zenity.io/)) - An independent source or public docs page confirming Zenity's SIEM, EDR and identity integrations beyond the vendor's own docs navigation; docs integration pages are partly behind login: none found on 13 Nov 2026 (checked: [docs.zenity.io](https://docs.zenity.io/)) - A published price. zenity.io/pricing returned 404 and the demo-only calls to action were the only buying path seen: none found on 12 Nov 2026 (checked: [zenity.io](https://zenity.io/pricing)) ## Position in each situation | Situation | Position | Total | Leader | | --- | --- | --- | --- | | [Coding agents](/best-for/coding-agents/) | 9th of 16 | 65.0 | Bay 75.5 | | [Regulated data](/best-for/regulated-data/) | 7th of 16 | 66.5 | Microsoft 79.0 | | [Lean team with EDR](/best-for/lean-team-edr/) | 10th of 16 (tied) | 62.5 | Bay 76.5 | | [Staff rollout](/best-for/staff-rollout/) | 7th of 16 | 65.5 | Microsoft 75.5 | | [Board in 30 days](/best-for/board-30-days/) | 11th of 16 | 60.5 | Bay 77.5 | Zenity under each situation's own weights. Tied marks an equal total; the order follows the tie-break, the higher score on the heaviest criterion, then name. ## Head-to-heads - [Pillar Security vs Zenity](/compare/pillar-security-vs-zenity/): Zenity 65.0 to Pillar Security 53.0 on the published weights **About Bay.** Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). In regulated data, Bay ranks 3rd of 16 at 70.5. In staff rollout, Bay ranks 2nd of 16 at 73.5. Bay is a young company that left stealth in 2026, publishes no price and has little independent testing behind its claims. More: [all 16 platform profiles](/platforms/), [every head-to-head](/compare/), [the six agent surfaces](/surfaces/) and [the six agent actions](/actions/). ## Sources - zenity.io/use-cases/agent-type/coding-personal-agents: https://zenity.io/use-cases/agent-type/coding-personal-agents (read 12 Nov 2026) - zenity.io/blog/coding-agent-attack-surface: https://zenity.io/blog/coding-agent-attack-surface (read 12 Nov 2026) - zenity.io/press-release/zenity-extends-ai-agent-security-and-governance-to-claude-enterprise: https://zenity.io/press-release/zenity-extends-ai-agent-security-and-governance-to-claude-enterprise (read 12 Nov 2026) - zenity.io/blog (checked, nothing found): https://zenity.io/blog (read 13 Nov 2026) - zenity.io/platform/mcp-security: https://zenity.io/platform/mcp-security (read 12 Nov 2026) - docs.zenity.io/user-guide/coding-agents (checked, nothing found): https://docs.zenity.io/user-guide/coding-agents (read 12 Nov 2026) - docs.zenity.io (checked, nothing found): https://docs.zenity.io/ (read 13 Nov 2026) - zenity.io: https://zenity.io/ (read 12 Nov 2026) - zenity.io/press-release/zenity-raises-125-million-to-secure-the-era-of-1-billion-ai-agents: https://zenity.io/press-release/zenity-raises-125-million-to-secure-the-era-of-1-billion-ai-agents (read 12 Nov 2026) - zenity.io/blog/zenity-for-claude-tag-securing-the-shared-agent-in-your-slack-workspace: https://zenity.io/blog/zenity-for-claude-tag-securing-the-shared-agent-in-your-slack-workspace (read 13 Nov 2026) - www.theregister.com/security/2026/09/24/salesforce-agentforce-vulns-allowed-0-click-crm-data-theft-anonymous-phishing/5298958: https://www.theregister.com/security/2026/09/24/salesforce-agentforce-vulns-allowed-0-click-crm-data-theft-anonymous-phishing/5298958 (read 12 Nov 2026) - www.securityweek.com/salesbleed-flaws-in-salesforce-agentforce-enabled-zero-click-data-exfiltration: https://www.securityweek.com/salesbleed-flaws-in-salesforce-agentforce-enabled-zero-click-data-exfiltration/ (read 12 Nov 2026) - www.techradar.com/pro/security/experts-warn-malicious-ai-skills-are-hitting-more-victims-than-ever-with-one-family-amassing-1-7-million-downloads: https://www.techradar.com/pro/security/experts-warn-malicious-ai-skills-are-hitting-more-victims-than-ever-with-one-family-amassing-1-7-million-downloads (read 12 Nov 2026) - zenity.io/press-release/zenity-labs-discloses-agentcorruption-a-chain-of-aws-agentcore-flaws: https://zenity.io/press-release/zenity-labs-discloses-agentcorruption-a-chain-of-aws-agentcore-flaws (read 13 Nov 2026) - zenity.io/pricing (checked, nothing found): https://zenity.io/pricing (read 12 Nov 2026) - www.calcalistech.com/ctechnews/article/rsdllsuy3: https://www.calcalistech.com/ctechnews/article/rsdllsuy3 (read 12 Nov 2026) - zenity.io/press-release/zenity-named-in-two-categories-in-the-2026-gartner-r-hype-cycle-tm-for-agentic-ai: https://zenity.io/press-release/zenity-named-in-two-categories-in-the-2026-gartner-r-hype-cycle-tm-for-agentic-ai (read 12 Nov 2026) - zenity.io/blog/gartner-market-shaper-ai-application-security: https://zenity.io/blog/gartner-market-shaper-ai-application-security (read 13 Nov 2026) - zenity.io/company: https://zenity.io/company (read 12 Nov 2026) - zenity.io/platform/ai-detection-and-response: https://zenity.io/platform/ai-detection-and-response (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/platforms/zenity/ --- # Knostic: 9th of 16 at 64.0, strongest on Transparency at 7 of 10 PLATFORM PROFILE, read 13 Nov 2026 **The short answer.** Knostic ranks 9th of 16 AI endpoint security tools at 64.0 of 100. Knostic stands out on Transparency at 7 of 10, and its lowest score is Beyond at 3 of 10. Knostic shows 4 of 6 agent surfaces in full and can stop 3 of 6 agent actions, as documented. ## Profile - **Owner:** Knostic Inc. (independent; no parent or acquirer named) - **Product:** Kirin (security for the agentic endpoint) and AgentMesh - **Status:** Active under this name, checked 13 Nov 2026 - **Category:** Agentic endpoint security: coding agents, MCP servers, extensions and skills - **Deployment:** Endpoint client (user-space, MDM-deployable) plus IDE extension and agent hooks - **Operating systems:** macOS, Windows, Linux (beta) - **Time to deploy:** "deploys in minutes" (vendor claim, FAQ) (the vendor's own claim) - **Price:** Kirin Free $0 (1 to 5 users), Team $15 per user per month, Enterprise custom; AgentMesh Free, Premium $15, Pro $850 per user per month (the vendor's own claim) - **Website:** [knostic.ai](https://www.knostic.ai/) Scores out of 10 (weight)Sees 257Stops 257Deployment 157Beyond 103Policy 106Transparency 107Proof 55 ## Strength and weakness **Strength:** Knostic documents Enforce-mode blocking of secret file reads, risky MCP servers, destructive commands and unsafe packages inside agents, with a published $15 per user Team price and free tier. **Weakness:** Shadow AI Discovery is visibility-only by design, supported-agent lists conflict between the FAQ and Kirin page, and none found on 12 Nov 2026 for named customers, analyst coverage or EDR integration. ## Scores: 64.0 of 100, 9th of 16 | Criterion | Weight | Score | Why this score | | --- | --- | --- | --- | | Sees agentic activity on the endpoint | 25 | 7 | The vendor names desktop, IDE and command-line agents and MCP servers, sees browser chatbots via network traffic, and open-source detection scripts were observed for one agent. Its own pages disagree on which agents are supported, and a second source for the full inventory was none found on 13 Nov 2026. | | Stops the action | 25 | 7 | The vendor says Enforce mode blocks secret file reads, destructive commands and risky MCP servers before they run; an observed open-source plugin blocks tool calls, but it is Alpha and for one agent. Shadow AI discovery is visibility-only, and a second-source test of the main product was none found on 13 Nov 2026. | | Deployment friction | 15 | 7 | The vendor publishes a light, MDM-pushed client for macOS and Windows (Linux beta) and claims deployment in minutes. All of it is vendor claim; an independent confirmation was none found on 13 Nov 2026. | | Beyond the endpoint | 10 | 3 | Coverage beyond coding agents is a plugin for one non-coding agent plus visibility-only discovery of browser chatbots. No browser, cloud workspace or SaaS agent enforcement under the same policy was found on 12 Nov 2026. | | Policy and governance | 10 | 6 | The vendor documents Off/Monitor/Enforce policies, audit logs, SIEM streaming and SSO, and an observed telemetry repo forwards to SIEM. No named EDR integration was found on 12 Nov 2026. | | Transparency | 10 | 7 | The vendor publishes list prices and a steady run of MCP research, and observed open-source repositories back the research. A readable public product documentation site was none found on 12 Nov 2026. | | Proof | 5 | 5 | Third-party press confirms an $11 million Series A and list awards, and the vendor says it was a Gartner Cool Vendor. No named customers were found on 12 Nov 2026. | Knostic: weighted total 64.0 of 100 on the published weights. ## What does Knostic see on an endpoint? Knostic documents 4 of 6 agent surfaces in full (desktop apps, IDE assistants, command-line agents and MCP servers) and 2 in part (browser extensions and SaaS copilots). Each cell with its note and source: [Knostic on the Agent Surface Map](/agent-surface-map/knostic/). ## Which agent actions can Knostic stop? Knostic documents stopping 3 of 6 agent actions before they complete (file read, shell command and credential access) and 3 in part (network call, data paste and package install). Each cell with its note and source: [Knostic on the Action Control Matrix](/action-control-matrix/knostic/). ## How much does Knostic cost? Knostic publishes a price: kirin Free $0 (1 to 5 users), Team $15 per user per month, Enterprise custom; AgentMesh Free, Premium $15, Pro $850 per user per month. (the vendor's own claim) ## Evidence behind the scores - Knostic's FAQ lists support for Claude Code CLI, Claude Desktop and Cowork, Cursor, Windsurf and VS Code with GitHub Copilot on macOS and Windows, with Linux in beta. (the vendor's own claim; [knostic.ai](https://www.knostic.ai/faq), read 12 Nov 2026) - Knostic's Kirin page lists Codex, JetBrains, Devin Desktop and Gemini CLI as supported, which differs from the FAQ that puts Codex and Gemini CLI on the roadmap. (the vendor's own claim; [knostic.ai](https://www.knostic.ai/kirin), read 12 Nov 2026) - Knostic's endpoint client watches network traffic to find shadow AI such as browser chatbots, local LLMs and unsanctioned LLM APIs. (the vendor's own claim; [knostic.ai](https://www.knostic.ai/faq), read 12 Nov 2026) - Knostic's Team tier adds cross-organization visibility into MCP servers, rules and hooks. (the vendor's own claim; [knostic.ai](https://www.knostic.ai/pricing), read 12 Nov 2026) - Knostic's open-source openclaw-detect repository ships detection scripts for macOS, Linux and Windows, built for MDM deployment (Jamf, Intune), that find the OpenClaw agent on managed devices. (observed in public documentation or code; [github.com](https://github.com/knostic/openclaw-detect), read 12 Nov 2026) - Knostic says every agent action is checked against policy before it runs and risky ones are blocked or sent to a human. (the vendor's own claim; [knostic.ai](https://www.knostic.ai), read 12 Nov 2026) - Knostic's FAQ says in Enforce mode Kirin blocks an agent's read of a file containing a secret or sanitizes the secret before it reaches the model. (the vendor's own claim; [knostic.ai](https://www.knostic.ai/faq), read 12 Nov 2026) - Knostic's FAQ says unverified or risky MCP servers are quarantined and removed from configuration so the agent cannot connect, pending admin review. (the vendor's own claim; [knostic.ai](https://www.knostic.ai/faq), read 12 Nov 2026) - Knostic's home page demo shows Kirin blocking a destructive database command from Claude Code before it ran. (the vendor's own claim; [knostic.ai](https://www.knostic.ai), read 12 Nov 2026) - Knostic's FAQ states enforcement is scoped to the agent layer and that Shadow AI Discovery is visibility-only by design. (the vendor's own claim; [knostic.ai](https://www.knostic.ai/faq), read 12 Nov 2026) - Knostic's open-source openclaw-shield plugin (labeled Alpha) hard-blocks dangerous tool calls and redacts secrets and PII from tool output in the OpenClaw agent. (observed in public documentation or code; [github.com](https://github.com/knostic/openclaw-shield), read 12 Nov 2026) - Knostic says Kirin deploys in minutes, pushed to endpoints directly or via MDM, starting in Monitor mode. (the vendor's own claim; [knostic.ai](https://www.knostic.ai/faq), read 12 Nov 2026) - Knostic says an IDE extension covers VS Code Copilot, Cursor and Windsurf, while Claude Code and Cowork require the endpoint client with agent hooks. (the vendor's own claim; [knostic.ai](https://www.knostic.ai/faq), read 12 Nov 2026) - Knostic's FAQ says supported agents run on macOS and Windows, with Linux in beta. (the vendor's own claim; [knostic.ai](https://www.knostic.ai/faq), read 12 Nov 2026) ## Checks that found nothing - An independent test, analyst note or trade-press write-up confirming which agents Kirin inventories, in Google and Bing News searches for Knostic with inventory, discovery, MCP and coding agents: none found on 13 Nov 2026 (checked: [news.google.com](https://news.google.com/rss/search?q=%22Knostic%22&hl=en-US&gl=US&ceid=US:en)) - An independent test or trade-press write-up confirming Kirin blocks agent actions before completion, in Google News searches for Knostic with block, runtime and enforcement: none found on 13 Nov 2026 (checked: [news.google.com](https://news.google.com/rss/search?q=%22Knostic%22&hl=en-US&gl=US&ceid=US:en)) - An independent source confirming Kirin deploy time or operating system coverage, in Google News searches for Knostic with agentless, deploy, macOS and Windows: none found on 13 Nov 2026 (checked: [news.google.com](https://news.google.com/rss/search?q=%22Knostic%22&hl=en-US&gl=US&ceid=US:en)) - A named EDR vendor integration (for example CrowdStrike or Defender) in Knostic's FAQ, Kirin page or pricing page: none found on 12 Nov 2026 (checked: [knostic.ai](https://www.knostic.ai/faq)) - A readable public product documentation site or changelog; docs.knostic.ai redirects to a GitBook space page and the website FAQ serves as documentation: none found on 12 Nov 2026 (checked: [docs.knostic.ai](https://docs.knostic.ai/)) - A third-party write-up of Knostic's published prices or of its MCP research posts (news coverage found was the OpenAnt scanner, funding and CB Insights list only), in Google and Bing News and the NVD and GitHub advisory databases: none found on 13 Nov 2026 (checked: [news.google.com](https://news.google.com/rss/search?q=%22Knostic%22&hl=en-US&gl=US&ceid=US:en)) - Named customers or an analyst report (Gartner, Forrester, IDC) naming Knostic on its home, Kirin or about pages: none found on 12 Nov 2026 (checked: [knostic.ai](https://www.knostic.ai)) ## Position in each situation | Situation | Position | Total | Leader | | --- | --- | --- | --- | | [Coding agents](/best-for/coding-agents/) | 8th of 16 | 66.0 | Bay 75.5 | | [Regulated data](/best-for/regulated-data/) | 11th of 16 | 60.5 | Microsoft 79.0 | | [Lean team with EDR](/best-for/lean-team-edr/) | 8th of 16 | 66.0 | Bay 76.5 | | [Staff rollout](/best-for/staff-rollout/) | 11th of 16 | 58.0 | Microsoft 75.5 | | [Board in 30 days](/best-for/board-30-days/) | 8th of 16 | 64.5 | Bay 77.5 | Knostic under each situation's own weights. Tied marks an equal total; the order follows the tie-break, the higher score on the heaviest criterion, then name. ## Head-to-heads - [Bay vs Knostic](/compare/bay-vs-knostic/): Bay 71.5 to Knostic 64.0 on the published weights - [Knostic vs Harmonic Security](/compare/knostic-vs-harmonic-security/): Harmonic Security 71.5 to Knostic 64.0 on the published weights - [Knostic vs Lasso Security](/compare/knostic-vs-lasso-security/): Knostic 64.0 to Lasso Security 53.0 on the published weights **About Bay.** Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). In regulated data, Bay ranks 3rd of 16 at 70.5. In staff rollout, Bay ranks 2nd of 16 at 73.5. Bay is a young company that left stealth in 2026, publishes no price and has little independent testing behind its claims. More: [all 16 platform profiles](/platforms/), [every head-to-head](/compare/), [the six agent surfaces](/surfaces/) and [the six agent actions](/actions/). ## Sources - www.knostic.ai/faq: https://www.knostic.ai/faq (read 12 Nov 2026) - www.knostic.ai/kirin: https://www.knostic.ai/kirin (read 12 Nov 2026) - www.knostic.ai/pricing: https://www.knostic.ai/pricing (read 12 Nov 2026) - github.com/knostic/openclaw-detect: https://github.com/knostic/openclaw-detect (read 12 Nov 2026) - news.google.com/rss/search?q=%22Knostic%22&hl=en-US&gl=US&ceid=US:en (checked, nothing found): https://news.google.com/rss/search?q=%22Knostic%22&hl=en-US&gl=US&ceid=US:en (read 13 Nov 2026) - www.knostic.ai: https://www.knostic.ai (read 12 Nov 2026) - github.com/knostic/openclaw-shield: https://github.com/knostic/openclaw-shield (read 12 Nov 2026) - siliconangle.com/2025/03/05/knostic-raises-11m-strengthen-enterprise-ai-security-need-know-access-controls: https://siliconangle.com/2025/03/05/knostic-raises-11m-strengthen-enterprise-ai-security-need-know-access-controls/ (read 12 Nov 2026) - github.com/knostic/openclaw-telemetry: https://github.com/knostic/openclaw-telemetry (read 12 Nov 2026) - www.knostic.ai/about: https://www.knostic.ai/about (read 12 Nov 2026) - www.knostic.ai/blog/when-an-mcp-tool-asks-for-your-github-token: https://www.knostic.ai/blog/when-an-mcp-tool-asks-for-your-github-token (read 13 Nov 2026) - github.com/knostic/AgentSonar: https://github.com/knostic/AgentSonar (read 12 Nov 2026) - github.com/knostic/OpenAnt: https://github.com/knostic/OpenAnt (read 12 Nov 2026) - docs.knostic.ai (checked, nothing found): https://docs.knostic.ai/ (read 12 Nov 2026) - www.securityweek.com/knostic-secures-11-million-to-rein-in-enterprise-ai-data-leakage-oversharing: https://www.securityweek.com/knostic-secures-11-million-to-rein-in-enterprise-ai-data-leakage-oversharing/ (read 12 Nov 2026) - finance.yahoo.com/sectors/technology/articles/knostic-named-2026-cb-insights-140000898.html: https://finance.yahoo.com/sectors/technology/articles/knostic-named-2026-cb-insights-140000898.html (read 12 Nov 2026) - www.knostic.ai/blog/gartner-cool-vendor-recognition: https://www.knostic.ai/blog/gartner-cool-vendor-recognition (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/platforms/knostic/ --- # SentinelOne: 10th of 16 at 62.0, strongest on Beyond at 7 of 10 PLATFORM PROFILE, read 13 Nov 2026 **The short answer.** SentinelOne ranks 10th of 16 AI endpoint security tools at 62.0 of 100. SentinelOne stands out on Beyond at 7 of 10, and its lowest score is Transparency at 4 of 10. SentinelOne shows 5 of 6 agent surfaces in full and can stop 1 of 6 agent actions, as documented. ## Profile - **Owner:** SentinelOne, Inc. (completed acquisition of Prompt Security on September 5, 2025, per its 10-Q) - **Product:** Prompt Security (SentinelOne) - **Status:** Active under this name, checked 13 Nov 2026 - **Category:** AI usage control and agent governance (browser extension plus endpoint AI agent, MCP gateway) from an EDR vendor - **Deployment:** Lightweight endpoint agent plus browser extensions, with an MCP gateway; Cowork VM agent enabled through device management (announced) - **Operating systems:** macOS, Windows - **Time to deploy:** none found on 12 Nov 2026 for the Prompt endpoint agent; SentinelOne says its narrower OneClaw tool deploys in minutes via Jamf, Intune, Kandji or Remote Ops - **Price:** none found on 12 Nov 2026 - **Website:** [prompt.security](https://prompt.security/) Scores out of 10 (weight)Sees 257Stops 257Deployment 155Beyond 107Policy 106Transparency 104Proof 55 ## Strength and weakness **Strength:** SentinelOne documents an MCP gateway with allow, block, filter or redact actions and runtime tool-call blocking, plus a lightweight agent and extension that discover shadow AI across browsers, IDEs and terminal assistants. **Weakness:** None found on 13 Nov 2026 for a published price, a public docs site or an OS list for the Prompt endpoint agent; the Cowork VM agent is announced, not shipped, and the AI Usage Control page describes coaching users rather than blocking. ## Scores: 62.0 of 100, 10th of 16 | Criterion | Weight | Score | Why this score | | --- | --- | --- | --- | | Sees agentic activity on the endpoint | 25 | 7 | The vendor says it discovers AI in browsers, IDEs, terminal assistants and MCP servers, with skills and connectors, and names code assistants it tracks; third-party coverage confirms MCP discovery and a browser extension was observed. The full endpoint inventory rests on vendor claims. | | Stops the action | 25 | 7 | The vendor says it blocks credential exfiltration, tool poisoning and malicious MCP servers before actions run, and third-party coverage reports it can stop agents sending data out. The newest blocking feature is early access, and the usage-control product also offers coaching instead of blocking. | | Deployment friction | 15 | 5 | The endpoint needs a browser extension plus an endpoint agent, macOS and Windows are announced for one integration, and the vendor says the usage-control product deploys in minutes. A full OS list for the main endpoint agent was none found on 12 Nov 2026. | | Beyond the endpoint | 10 | 7 | The vendor says coverage spans browsers, desktop apps, APIs and SaaS agent platforms, and that policy is enforced consistently across browser, desktop and embedded copilots. All of it is vendor claim. | | Policy and governance | 10 | 6 | The vendor offers allow and block rules by user, agent and action, a full audit trail, an EDR execution graph and a customer-reported tie to identity systems. A named SIEM or identity-provider integration was none found on 12 Nov 2026. | | Transparency | 10 | 4 | The vendor publishes agent-risk research and free tools, and its public open-source repositories were observed. A published price and public documentation were none found on 12 Nov 2026. | | Proof | 5 | 5 | The vendor cites Gartner, Latio and an analyst-firm recognition, gives parent-company figures, and now names customers Carnival Corporation and 10x Banking in case studies. Product-specific funding and third-party confirmation of the customers were none found on 13 Nov 2026. | SentinelOne: weighted total 62.0 of 100 on the published weights. ## What does SentinelOne see on an endpoint? SentinelOne documents 5 of 6 agent surfaces in full (desktop apps, IDE assistants, command-line agents, MCP servers and browser extensions) and 1 in part (SaaS copilots). Each cell with its note and source: [SentinelOne on the Agent Surface Map](/agent-surface-map/sentinelone/). ## Which agent actions can SentinelOne stop? SentinelOne documents stopping 1 of 6 agent actions before they complete (data paste) and 3 in part (shell command, network call and credential access). Nothing was found for file read and package install. Each cell with its note and source: [SentinelOne on the Action Control Matrix](/action-control-matrix/sentinelone/). ## How much does SentinelOne cost? SentinelOne publishes no list price. None found on 12 Nov 2026. Price counts toward no score; deployment friction is scored under Deployment. ## Evidence behind the scores - SentinelOne says Prompt's lightweight agent and browser extensions discover sanctioned and shadow AI across browsers, desktop IDEs and terminal-based assistants. (the vendor's own claim; [sentinelone.com](https://www.sentinelone.com/blog/a-new-chapter-for-ai-and-cybersecurity-sentinelone-acquires-prompt-security/), read 12 Nov 2026) - SentinelOne says Prompt for Agentic AI Security surfaces every MCP server, sanctioned or shadow. (the vendor's own claim; [sentinelone.com](https://www.sentinelone.com/blog/prompt-security-for-agentic-ai/), read 12 Nov 2026) - Prompt Security's engineers describe a dedicated endpoint AI agent that detects new AI agents and understands active skills and connectors. (the vendor's own claim; [prompt.security](https://prompt.security/blog/the-key-layer-in-ai-security-browser-and-endpoint-sensors), read 12 Nov 2026) - SentinelOne says ADR inventories every agent, tool, skill, connector and MCP server across endpoints, SaaS and cloud with the permissions each holds, and extends to Salesforce Agentforce, Copilot Studio, ChatGPT workspace agents and Claude managed agents. (the vendor's own claim; [sentinelone.com](https://www.sentinelone.com/press/sentinelone-launches-singularity-agentic-detection-and-response-to-stop-rogue-and-malicious-ai-agents-at-machine-speed/), read 12 Nov 2026) - SiliconANGLE reports Prompt AI Agent Security can automatically identify the MCP servers a company’s AI agents use. (third-party source; [siliconangle.com](https://siliconangle.com/2026/03/23/sentinelone-snyk-introduce-new-tools-securing-ai-agents/), read 12 Nov 2026) - The Chrome Web Store lists a Prompt Security Browser Extension from prompt.security, rated 4.8, described as preventing sensitive information notice to GenAI tools. (observed in public documentation or code; [chromewebstore.google.com](https://chromewebstore.google.com/search/prompt%20security), read 12 Nov 2026) - SentinelOne AI Usage Control page names GitHub Copilot, Cursor and Claude Code among the AI code assistants it tracks. (the vendor's own claim; [sentinelone.com](https://www.sentinelone.com/platform/prompt-security-ai-usage-control/), read 13 Nov 2026) - SentinelOne says its MCP gateway scores each server and enforces allow, block, filter or redact actions. (the vendor's own claim; [sentinelone.com](https://www.sentinelone.com/blog/a-new-chapter-for-ai-and-cybersecurity-sentinelone-acquires-prompt-security/), read 12 Nov 2026) - SentinelOne says Prompt for Agentic AI Security blocks prompt injection at tool-call time. (the vendor's own claim; [sentinelone.com](https://www.sentinelone.com/blog/prompt-security-for-agentic-ai/), read 12 Nov 2026) - SentinelOne announced Singularity Agentic Detection and Response (ADR) on October 21, 2026, in early access, saying it blocks prompt injection, jailbreaks, credential exfiltration and tool poisoning before the action executes. (the vendor's own claim; [sentinelone.com](https://www.sentinelone.com/press/sentinelone-launches-singularity-agentic-detection-and-response-to-stop-rogue-and-malicious-ai-agents-at-machine-speed/), read 12 Nov 2026) - SentinelOne says ADR gives allow and block controls for which agents run, who uses them and what they can reach, with approval workflows for high-risk tools, and can contain one agent mid-action while the rest of the workflow runs. (the vendor's own claim; [sentinelone.com](https://www.sentinelone.com/press/sentinelone-launches-singularity-agentic-detection-and-response-to-stop-rogue-and-malicious-ai-agents-at-machine-speed/), read 12 Nov 2026) - SiliconANGLE reports Prompt AI Agent Security (March 2026) identifies MCP servers used by AI agents, applies guardrails to them, and can stop agents sending business data to a third-party app or reaching unauthorized internal systems. (third-party source; [siliconangle.com](https://siliconangle.com/2026/03/23/sentinelone-snyk-introduce-new-tools-securing-ai-agents/), read 12 Nov 2026) - SentinelOne chief product officer Chris Corde told CRN that Singularity ADR combines Prompt Security technology with the endpoint (EDR) agent to trace an agent action back to the originating prompt (SentinelOne executive speaking in trade press). (third-party source; [crn.com](https://www.crn.com/news/security/2026/sentinelone-s-chris-corde-on-why-traditional-edr-is-flying-blind-on-ai-agents), read 13 Nov 2026) - Limits the blocking claim: the AI Usage Control page describes coaching users in the moment instead of blocking as a mode of enforcement. (the vendor's own claim; [sentinelone.com](https://www.sentinelone.com/platform/prompt-security-ai-usage-control/), read 13 Nov 2026) ## Checks that found nothing - A published time to deploy or OS list for the Prompt Security endpoint agent; SentinelOne's Prompt product pages returned a Vercel security checkpoint to curl and reader proxies: none found on 12 Nov 2026 (checked: [sentinelone.com](https://www.sentinelone.com/platform/prompt-security-ai-usage-control/)) - Documented SIEM or identity-provider integrations specific to Prompt Security on pages read (SentinelOne product pages blocked automated reads): none found on 12 Nov 2026 (checked: [sentinelone.com](https://www.sentinelone.com/platform/prompt-security-agentic-ai-security/)) - A published price for Prompt Security; prompt.security/pricing returned Not Found and SentinelOne's pricing and packages page returned a security checkpoint: none found on 12 Nov 2026 (checked: [prompt.security](https://prompt.security/pricing)) - A public documentation site for Prompt Security; docs.prompt.security returned an empty response: none found on 12 Nov 2026 (checked: [docs.prompt.security](https://docs.prompt.security/)) ## Position in each situation | Situation | Position | Total | Leader | | --- | --- | --- | --- | | [Coding agents](/best-for/coding-agents/) | 10th of 16 | 63.5 | Bay 75.5 | | [Regulated data](/best-for/regulated-data/) | 10th of 16 (tied) | 64.0 | Microsoft 79.0 | | [Lean team with EDR](/best-for/lean-team-edr/) | 11th of 16 | 61.5 | Bay 76.5 | | [Staff rollout](/best-for/staff-rollout/) | 8th of 16 (tied) | 63.5 | Microsoft 75.5 | | [Board in 30 days](/best-for/board-30-days/) | 10th of 16 | 61.0 | Bay 77.5 | SentinelOne under each situation's own weights. Tied marks an equal total; the order follows the tie-break, the higher score on the heaviest criterion, then name. ## Head-to-heads - [Microsoft vs SentinelOne](/compare/microsoft-vs-sentinelone/): Microsoft 75.0 to SentinelOne 62.0 on the published weights - [WitnessAI vs SentinelOne](/compare/witnessai-vs-sentinelone/): SentinelOne 62.0 to WitnessAI 61.0 on the published weights **About Bay.** Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). In regulated data, Bay ranks 3rd of 16 at 70.5. In staff rollout, Bay ranks 2nd of 16 at 73.5. Bay is a young company that left stealth in 2026, publishes no price and has little independent testing behind its claims. More: [all 16 platform profiles](/platforms/), [every head-to-head](/compare/), [the six agent surfaces](/surfaces/) and [the six agent actions](/actions/). ## Sources - www.sentinelone.com/blog/a-new-chapter-for-ai-and-cybersecurity-sentinelone-acquires-prompt-security: https://www.sentinelone.com/blog/a-new-chapter-for-ai-and-cybersecurity-sentinelone-acquires-prompt-security/ (read 12 Nov 2026) - www.sentinelone.com/blog/prompt-security-for-agentic-ai: https://www.sentinelone.com/blog/prompt-security-for-agentic-ai/ (read 12 Nov 2026) - prompt.security/blog/the-key-layer-in-ai-security-browser-and-endpoint-sensors: https://prompt.security/blog/the-key-layer-in-ai-security-browser-and-endpoint-sensors (read 12 Nov 2026) - www.sentinelone.com/press/sentinelone-launches-singularity-agentic-detection-and-response-to-stop-rogue-and-malicious-ai-agents-at-machine-speed: https://www.sentinelone.com/press/sentinelone-launches-singularity-agentic-detection-and-response-to-stop-rogue-and-malicious-ai-agents-at-machine-speed/ (read 12 Nov 2026) - siliconangle.com/2026/03/23/sentinelone-snyk-introduce-new-tools-securing-ai-agents: https://siliconangle.com/2026/03/23/sentinelone-snyk-introduce-new-tools-securing-ai-agents/ (read 12 Nov 2026) - chromewebstore.google.com/search/prompt%20security: https://chromewebstore.google.com/search/prompt%20security (read 12 Nov 2026) - www.sentinelone.com/platform/prompt-security-ai-usage-control: https://www.sentinelone.com/platform/prompt-security-ai-usage-control/ (read 13 Nov 2026) - www.crn.com/news/security/2026/sentinelone-s-chris-corde-on-why-traditional-edr-is-flying-blind-on-ai-agents: https://www.crn.com/news/security/2026/sentinelone-s-chris-corde-on-why-traditional-edr-is-flying-blind-on-ai-agents (read 13 Nov 2026) - www.sentinelone.com/press/sentinelone-expands-ai-security-coverage-to-claude-cowork: https://www.sentinelone.com/press/sentinelone-expands-ai-security-coverage-to-claude-cowork/ (read 12 Nov 2026) - www.sentinelone.com/blog/oneclaw-discovery-and-observability-for-the-agentic-era: https://www.sentinelone.com/blog/oneclaw-discovery-and-observability-for-the-agentic-era/ (read 12 Nov 2026) - www.sentinelone.com/customers/10xbanking: https://www.sentinelone.com/customers/10xbanking/ (read 13 Nov 2026) - www.sentinelone.com/press/sentinelone-to-acquire-prompt-security-to-advance-genai-security: https://www.sentinelone.com/press/sentinelone-to-acquire-prompt-security-to-advance-genai-security (read 12 Nov 2026) - www.sentinelone.com/press/sentinelone-named-an-ai-security-platform-leader-in-latio-2026-ai-security-market-report: https://www.sentinelone.com/press/sentinelone-named-an-ai-security-platform-leader-in-latio-2026-ai-security-market-report/ (read 12 Nov 2026) - www.sentinelone.com/platform/prompt-security-agentic-ai-security (checked, nothing found): https://www.sentinelone.com/platform/prompt-security-agentic-ai-security/ (read 12 Nov 2026) - prompt.security/blog/when-your-plugin-starts-picking-your-dependencies-marketplace-skills-and-dependency-hijack-in-claude-code: https://prompt.security/blog/when-your-plugin-starts-picking-your-dependencies-marketplace-skills-and-dependency-hijack-in-claude-code (read 12 Nov 2026) - prompt.security/blog: https://prompt.security/blog (read 12 Nov 2026) - github.com/prompt-security/clawsec: https://github.com/prompt-security/clawsec (read 12 Nov 2026) - github.com/prompt-security: https://github.com/prompt-security (read 12 Nov 2026) - prompt.security/blog/when-ai-agents-become-the-supply-chain-hidden-control-planes-in-agentic-systems: https://prompt.security/blog/when-ai-agents-become-the-supply-chain-hidden-control-planes-in-agentic-systems (read 13 Nov 2026) - prompt.security/pricing (checked, nothing found): https://prompt.security/pricing (read 12 Nov 2026) - docs.prompt.security (checked, nothing found): https://docs.prompt.security/ (read 12 Nov 2026) - prompt.security/blog/prompt-security-named-as-a-2025-gartner-cool-vendor-in-ai-security: https://prompt.security/blog/prompt-security-named-as-a-2025-gartner-cool-vendor-in-ai-security (read 12 Nov 2026) - www.sentinelone.com/press/sentinelone-scores-top-marks-in-unified-agentic-defense-platform-evaluation: https://www.sentinelone.com/press/sentinelone-scores-top-marks-in-unified-agentic-defense-platform-evaluation/ (read 12 Nov 2026) - www.sentinelone.com/customers/carnival: https://www.sentinelone.com/customers/carnival/ (read 13 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/platforms/sentinelone/ --- # WitnessAI: 11th of 16 at 61.0, strongest on Sees at 8 of 10 PLATFORM PROFILE, read 13 Nov 2026 **The short answer.** WitnessAI ranks 11th of 16 AI endpoint security tools at 61.0 of 100. WitnessAI stands out on Sees at 8 of 10, and its lowest score is Transparency at 3 of 10. WitnessAI shows 3 of 6 agent surfaces in full and can stop 2 of 6 agent actions, as documented. ## Profile - **Owner:** WitnessAI (independent, venture-backed; no acquirer found) - **Product:** WitnessAI Unified AI Security and Governance Platform (Observe, Control, Protect, Attack; WitnessAI for Agents) - **Status:** Active under this name, checked 13 Nov 2026 - **Category:** Network-level AI security and governance platform for employees, applications and agents, with MCP and tool-access enforcement - **Deployment:** Network-level, in the traffic path; no endpoint client or browser extension required for visibility (vendor) - **Operating systems:** none found on 12 Nov 2026 - **Time to deploy:** none found on 12 Nov 2026 - **Price:** none found on 12 Nov 2026 (AWS Marketplace lists the product with no price shown) - **Website:** [witness.ai](https://witness.ai/) Scores out of 10 (weight)Sees 258Stops 256Deployment 155Beyond 106Policy 106Transparency 103Proof 57 ## Strength and weakness **Strength:** WitnessAI documents inline blocking of unapproved MCP servers and tool calls across IDEs, chat apps and custom agents, with audit records per blocked call and one policy for humans and agents. **Weakness:** WitnessAI works at the network layer and says it needs no endpoint client; none found on 12 Nov 2026: an OS list, deploy time, price, public docs or named EDR and SIEM integrations, and its own text says protection depends on traffic passing through its controls. ## Scores: 61.0 of 100, 11th of 16 | Criterion | Weight | Score | Why this score | | --- | --- | --- | --- | | Sees agentic activity on the endpoint | 25 | 8 | The vendor names desktop clients, IDEs, coding agents, plugins and MCP servers, and third-party coverage confirms discovery of desktop plugins, IDE extensions, local frameworks and MCP servers, plus an MCP catalog. Discovery works from network traffic, and an on-device inventory of skills or packages was none found on 12 Nov 2026. | | Stops the action | 25 | 6 | The vendor says it blocks unapproved MCP servers and tool calls inline at the network, and third-party coverage confirms prompt-injection blocking and credential tokenizing. Enforcement depends on traffic passing through it, and endpoint blocking of shell or file actions was none found on 12 Nov 2026. | | Deployment friction | 15 | 5 | The vendor deploys at the network level or through lightweight APIs with no endpoint client, and its CEO says one unnamed large customer was running in 3 days. An operating-system list was none found on 12 Nov 2026, and the deploy time is a single vendor anecdote. | | Beyond the endpoint | 10 | 6 | The vendor says one policy covers humans, IDEs, chat apps and custom cloud agents, with orchestration-layer protection for custom agents. All of it is vendor claim, and browser-specific or named SaaS-agent coverage was none found on 13 Nov 2026. | | Policy and governance | 10 | 6 | The vendor records user, agent, tool and rule for each block and ties identity to the IdP; third-party coverage confirms one policy and audit trail. A named EDR or SIEM integration was none found on 12 Nov 2026. | | Transparency | 10 | 3 | The vendor publishes security research on its blog and a self-published guardrail benchmark figure. Docs are password-gated and a published price was none found on 12 Nov 2026. | | Proof | 5 | 7 | Third-party press confirms more than $85 million raised, and the vendor names a customer, a reseller, an alliance membership and an analyst market map. An independent analyst rating or a customer named by a third party was none found on 13 Nov 2026. | WitnessAI: weighted total 61.0 of 100 on the published weights. ## What does WitnessAI see on an endpoint? WitnessAI documents 3 of 6 agent surfaces in full (desktop apps, IDE assistants and MCP servers) and 3 in part (command-line agents, browser extensions and SaaS copilots). Each cell with its note and source: [WitnessAI on the Agent Surface Map](/agent-surface-map/witnessai/). ## Which agent actions can WitnessAI stop? WitnessAI documents stopping 2 of 6 agent actions before they complete (network call and data paste) and 2 in part (shell command and credential access). Nothing was found for file read and package install. Each cell with its note and source: [WitnessAI on the Action Control Matrix](/action-control-matrix/witnessai/). ## How much does WitnessAI cost? WitnessAI publishes no list price. None found on 12 Nov 2026 (AWS Marketplace lists the product with no price shown) Price counts toward no score; deployment friction is scored under Deployment. ## Evidence behind the scores - WitnessAI says it finds and catalogs thousands of AI applications, agents and MCP servers. (the vendor's own claim; [witness.ai](https://witness.ai/), read 12 Nov 2026) - WitnessAI says it captures prompts, tool calls and shell executions across supported Cursor, Claude Code and Codex workflows. (the vendor's own claim; [witness.ai](https://witness.ai/for-agents/), read 12 Nov 2026) - WitnessAI says it shows which agentic plugins employees install and which MCP servers they reach. (the vendor's own claim; [witness.ai](https://witness.ai/for-employees/), read 12 Nov 2026) - SiliconANGLE reports WitnessAI's Agentic Security discovers agent activity in Claude Desktop and plugins, ChatGPT plugins, VS Code AI extensions and local frameworks such as LangChain and CrewAI. (third-party source; [siliconangle.com](https://siliconangle.com/2026/01/13/witnessai-debuts-agentic-security-enterprises-deploy-autonomous-ai-agents/), read 12 Nov 2026) - SiliconANGLE reports the platform tells chat sessions from agentic sessions by analyzing tool usage in traffic payloads and maps public and private MCP servers being accessed. (third-party source; [siliconangle.com](https://siliconangle.com/2026/01/13/witnessai-debuts-agentic-security-enterprises-deploy-autonomous-ai-agents/), read 12 Nov 2026) - The June 2026 newswire release says Agentic Control discovers agents across IDEs, applications, agent frameworks and custom cloud agents, and identifies the MCP servers and tools each reaches. (the vendor's own claim; [prnewswire.com](https://www.prnewswire.com/news-releases/witnessai-introduces-agentic-control-to-secure-and-govern-ai-agents-and-mcp-servers-302802606), read 12 Nov 2026) - Help Net Security reports (from WitnessAI's June 2026 release) a new MCP Catalog that scores known tools against OWASP and CVE risk classes before approval. (third-party source; [helpnetsecurity.com](https://www.helpnetsecurity.com/2026/06/17/witnessai-agentic-control/), read 13 Nov 2026) - In the Pulse 2.0 interview the CEO says the platform inventories AI plugins, local agent environments and MCP server connections across Claude Desktop, VSCode, ChatGPT and developer machines. (third-party source; [pulse2.com](https://pulse2.com/witnessai-profile-rick-caccia-interview/), read 13 Nov 2026) - WitnessAI says it blocks an unapproved MCP server or tool call inline before it executes. (the vendor's own claim; [witness.ai](https://witness.ai/blog/witnessai-in-sacrs-arise-market-map-why-a-unified-ai-control-plane-matters/), read 12 Nov 2026) - WitnessAI says one organization-wide approved-tool list of MCP servers and tools is enforced at the network for every agent across every IDE, chat app and custom agent. (the vendor's own claim; [witness.ai](https://witness.ai/product/), read 12 Nov 2026) - SecurityWeek reports WitnessAI blocks multi-turn and prompt injection attacks before they reach AI agents. (third-party source; [securityweek.com](https://www.securityweek.com/witnessai-raises-58-million-for-ai-security-platform/), read 12 Nov 2026) - SiliconANGLE reports prompts are scanned before execution to detect and block prompt injection, jailbreaking and role-playing attacks, with credentials and PII tokenized in real time. (third-party source; [siliconangle.com](https://siliconangle.com/2026/01/13/witnessai-debuts-agentic-security-enterprises-deploy-autonomous-ai-agents/), read 12 Nov 2026) - WitnessAI's own release says its enforcement is real-time at the network layer over the tools and MCP servers agents can access (network enforcement, not an endpoint hook on shell or file actions). (the vendor's own claim; [prnewswire.com](https://www.prnewswire.com/news-releases/witnessai-introduces-agentic-control-to-secure-and-govern-ai-agents-and-mcp-servers-302802606), read 12 Nov 2026) - Help Net Security repeats WitnessAI's description of Agentic Control as real-time network enforcement over the tools and MCP servers agents can access (a trade-press copy of the vendor release, not independent testing). (third-party source; [helpnetsecurity.com](https://www.helpnetsecurity.com/2026/06/17/witnessai-agentic-control/), read 13 Nov 2026) ## Checks that found nothing - Supported operating systems, an endpoint component, or a time-to-deploy figure on the product, observe, control or agents pages: none found on 12 Nov 2026 (checked: [witness.ai](https://witness.ai/product/)) - A second (third-party or observed) source for WitnessAI covering browsers, cloud workspaces or SaaS-hosted agents in one policy; the agents page and news results repeat vendor wording only: none found on 13 Nov 2026 (checked: [witness.ai](https://witness.ai/for-agents/)) - Named EDR or SIEM integrations on the product, observe, control or agents pages: none found on 12 Nov 2026 (checked: [witness.ai](https://witness.ai/product/)) - A published price; witness.ai/pricing returns a page-not-found error and product pages route to a demo: none found on 12 Nov 2026 (checked: [witness.ai](https://witness.ai/pricing/)) - An independent analyst report (Gartner, Forrester, IDC) rating WitnessAI or a customer named by a third party in Google News and Bing News results; the Software Analyst ARISE page returned HTTP 403 to this fetch: none found on 13 Nov 2026 (checked: [news.google.com](https://news.google.com/rss/search?q=%22WitnessAI%22+(customer+OR+funding+OR+Gartner+OR+Forrester+OR+analyst)&hl=en-US&gl=US&ceid=US:en)) ## Position in each situation | Situation | Position | Total | Leader | | --- | --- | --- | --- | | [Coding agents](/best-for/coding-agents/) | 11th of 16 | 62.5 | Bay 75.5 | | [Regulated data](/best-for/regulated-data/) | 8th of 16 | 64.5 | Microsoft 79.0 | | [Lean team with EDR](/best-for/lean-team-edr/) | 9th of 16 (tied) | 62.5 | Bay 76.5 | | [Staff rollout](/best-for/staff-rollout/) | 10th of 16 | 61.5 | Microsoft 75.5 | | [Board in 30 days](/best-for/board-30-days/) | 9th of 16 | 64.0 | Bay 77.5 | WitnessAI under each situation's own weights. Tied marks an equal total; the order follows the tie-break, the higher score on the heaviest criterion, then name. ## Head-to-heads - [Bay vs WitnessAI](/compare/bay-vs-witnessai/): Bay 71.5 to WitnessAI 61.0 on the published weights - [WitnessAI vs SentinelOne](/compare/witnessai-vs-sentinelone/): SentinelOne 62.0 to WitnessAI 61.0 on the published weights **About Bay.** Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). In regulated data, Bay ranks 3rd of 16 at 70.5. In staff rollout, Bay ranks 2nd of 16 at 73.5. Bay is a young company that left stealth in 2026, publishes no price and has little independent testing behind its claims. More: [all 16 platform profiles](/platforms/), [every head-to-head](/compare/), [the six agent surfaces](/surfaces/) and [the six agent actions](/actions/). ## Sources - witness.ai: https://witness.ai/ (read 12 Nov 2026) - witness.ai/for-agents: https://witness.ai/for-agents/ (read 12 Nov 2026) - witness.ai/for-employees: https://witness.ai/for-employees/ (read 12 Nov 2026) - siliconangle.com/2026/01/13/witnessai-debuts-agentic-security-enterprises-deploy-autonomous-ai-agents: https://siliconangle.com/2026/01/13/witnessai-debuts-agentic-security-enterprises-deploy-autonomous-ai-agents/ (read 12 Nov 2026) - www.prnewswire.com/news-releases/witnessai-introduces-agentic-control-to-secure-and-govern-ai-agents-and-mcp-servers-302802606: https://www.prnewswire.com/news-releases/witnessai-introduces-agentic-control-to-secure-and-govern-ai-agents-and-mcp-servers-302802606 (read 12 Nov 2026) - www.helpnetsecurity.com/2026/06/17/witnessai-agentic-control: https://www.helpnetsecurity.com/2026/06/17/witnessai-agentic-control/ (read 13 Nov 2026) - pulse2.com/witnessai-profile-rick-caccia-interview: https://pulse2.com/witnessai-profile-rick-caccia-interview/ (read 13 Nov 2026) - witness.ai/blog/witnessai-in-sacrs-arise-market-map-why-a-unified-ai-control-plane-matters: https://witness.ai/blog/witnessai-in-sacrs-arise-market-map-why-a-unified-ai-control-plane-matters/ (read 12 Nov 2026) - witness.ai/product: https://witness.ai/product/ (read 12 Nov 2026) - www.securityweek.com/witnessai-raises-58-million-for-ai-security-platform: https://www.securityweek.com/witnessai-raises-58-million-for-ai-security-platform/ (read 12 Nov 2026) - witness.ai/resources/witnessai-announces-integration-with-f5-big-ip-to-deliver-visibility-and-control-over-encrypted-genai-activity: https://witness.ai/resources/witnessai-announces-integration-with-f5-big-ip-to-deliver-visibility-and-control-over-encrypted-genai-activity/ (read 12 Nov 2026) - witness.ai/control: https://witness.ai/control/ (read 12 Nov 2026) - witness.ai/resources/witnessai-named-a-founding-member-of-servicenow-control-tower-alliance: https://witness.ai/resources/witnessai-named-a-founding-member-of-servicenow-control-tower-alliance/ (read 13 Nov 2026) - witness.ai/blog/malicious-mcp-servers: https://witness.ai/blog/malicious-mcp-servers/ (read 12 Nov 2026) - docs.witness.ai: https://docs.witness.ai (read 12 Nov 2026) - witness.ai/pricing (checked, nothing found): https://witness.ai/pricing/ (read 12 Nov 2026) - witness.ai/resources/incomm-payments: https://witness.ai/resources/incomm-payments/ (read 12 Nov 2026) - witness.ai/resources/witnessai-and-ntt-data-japan-partner-to-accelerate-secure-ai-adoption-for-japanese-enterprises: https://witness.ai/resources/witnessai-and-ntt-data-japan-partner-to-accelerate-secure-ai-adoption-for-japanese-enterprises/ (read 12 Nov 2026) - news.google.com/rss/search?q=%22WitnessAI%22+(customer+OR+funding+OR+Gartner+OR+Forrester+OR+analyst)&hl=en-US&gl=US&ceid=US:en (checked, nothing found): https://news.google.com/rss/search?q=%22WitnessAI%22+(customer+OR+funding+OR+Gartner+OR+Forrester+OR+analyst)&hl=en-US&gl=US&ceid=US:en (read 13 Nov 2026) - witness.ai/protect: https://witness.ai/protect/ (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/platforms/witnessai/ --- # Koi: 12th of 16 at 54.5, strongest on Transparency at 6 of 10 PLATFORM PROFILE, read 13 Nov 2026 **The short answer.** Koi ranks 12th of 16 AI endpoint security tools at 54.5 of 100. Koi stands out on Transparency at 6 of 10, and its lowest score is Beyond at 2 of 10. Koi shows 4 of 6 agent surfaces in full and can stop 0 of 6 agent actions, as documented. ## Profile - **Listed as:** Koi (Cortex Agentic Endpoint Security) - **Owner:** Palo Alto Networks, Inc. - **Product:** Cortex Agentic Endpoint Security (formerly Koi, acquired by Palo Alto Networks) - **Status:** Active under this name, checked 13 Nov 2026 - **Category:** Agentic endpoint security: discovery and governance of AI agents, extensions, plugins and MCP servers on employee endpoints - **Deployment:** Endpoint security solution sold standalone (its own AES tenant and license) or as a core module of Cortex XDR; a Cortex Targeted Sensor runs it beside third-party EDR (vendor docs) - **Operating systems:** Windows, macOS, Linux - **Time to deploy:** none found on 12 Nov 2026 - **Price:** none found on 12 Nov 2026 - **Website:** [paloaltonetworks.com](https://www.paloaltonetworks.com/cortex/agentic-endpoint-security) Scores out of 10 (weight)Sees 257Stops 255Deployment 156Beyond 102Policy 104Transparency 106Proof 57 ## Strength and weakness **Strength:** Palo Alto Networks documents inventory of IDE plugins, browser extensions and MCP servers on endpoints, plus real-time blocking of risky downloads and agent actions, with standalone or Cortex XDR deployment. **Weakness:** None found on 13 Nov 2026: public detail on which agent action types are blocked, deployment time, price or SIEM and identity-provider integrations, or an independent analyst rating; a suit over a Koi research report is pending, allegations only. ## Scores: 54.5 of 100, 12th of 16 | Criterion | Weight | Score | Why this score | | --- | --- | --- | --- | | Sees agentic activity on the endpoint | 25 | 7 | The vendor lists browser extensions, IDE plugins, MCP servers, packages and local models across endpoints with a real-time inventory, and third-party coverage describes software inventory and relays the vendor naming Claude Code, Codex CLI and OpenClaw among tools covered by the combined solution. A second source confirming the on-device agent inventory detail was none found on 13 Nov 2026. | | Stops the action | 25 | 5 | Observed docs say runtime protection intercepts policy-violating AI coding agent actions before they execute, trade press reports OS-level interception of prompts and responses, and application control blocks untrusted apps. The blocked agent action types are not named, which was none found on 13 Nov 2026. | | Deployment friction | 15 | 6 | Observed release notes list Windows, macOS and Linux, and observed docs and trade press describe a targeted sensor that runs beside existing third-party EDR under a separate license. A time-to-deploy figure was none found on 13 Nov 2026. | | Beyond the endpoint | 10 | 2 | The vendor plans integration with a wider platform and has a software supply-chain gateway. No browser, cloud workspace or SaaS-agent coverage under one policy was found on 12 Nov 2026. | | Policy and governance | 10 | 4 | The vendor ships out-of-the-box agent policies, observed docs show a session-centric audit trail of every coding agent action, and trade press reports policies mapped to MITRE ATLAS. SIEM export or identity-provider integration was none found on 13 Nov 2026. | | Transparency | 10 | 6 | The vendor's docs are public and third-party press covers its research, though one report is the subject of a defamation suit and was edited to remove the claimant. A published price was none found on 12 Nov 2026. | | Proof | 5 | 7 | Third-party press confirms $48 million raised and an acquisition by a large vendor, though reported deal values differ ($400 million against $231 million), and customer logos were observed. An independent analyst placement was none found on 13 Nov 2026. | Koi: weighted total 54.5 of 100 on the published weights. ## What does Koi see on an endpoint? Koi documents 4 of 6 agent surfaces in full (IDE assistants, command-line agents, MCP servers and browser extensions) and 1 in part (desktop apps). Nothing was found for SaaS copilots. Each cell with its note and source: [Koi on the Agent Surface Map](/agent-surface-map/koi/). ## Which agent actions can Koi stop? Koi documents stopping 0 of 6 agent actions before they complete and 4 in part (shell command, network call, data paste and package install). Nothing was found for file read and credential access. Each cell with its note and source: [Koi on the Action Control Matrix](/action-control-matrix/koi/). ## How much does Koi cost? Koi publishes no list price. None found on 12 Nov 2026. Price counts toward no score; deployment friction is scored under Deployment. ## Evidence behind the scores - Palo Alto Networks says AES discovers shadow AI, browser extensions, IDE plugins, MCP servers and other unmanaged software across endpoints. (the vendor's own claim; [paloaltonetworks.com](https://www.paloaltonetworks.com/resources/datasheets/cortex-agentic-endpoint-security-at-a-glance), read 12 Nov 2026) - The tech brief lists the endpoint items AES governs: user-deployed AI applications, packages, browser extensions, local LLM models and MCP servers. (the vendor's own claim; [paloaltonetworks.com](https://www.paloaltonetworks.com/resources/techbriefs/koi-agentic-endpoint-security), read 12 Nov 2026) - Palo Alto Networks names Claude Code and OpenClaw as the agent tools expanding the endpoint attack surface that Koi technology addresses. (the vendor's own claim; [paloaltonetworks.com](https://www.paloaltonetworks.com/company/press/2026/palo-alto-networks-completes-acquisition-of-koi-to-secure-the-agentic-endpoint), read 12 Nov 2026) - SecurityWeek describes the Koi platform as protecting applications, code, OS packages, extensions, AI models, AI agents and containers. (third-party source; [securityweek.com](https://www.securityweek.com/palo-alto-networks-to-acquire-koi-in-reported-400-million-transaction/), read 12 Nov 2026) - SecurityWeek describes the Koi platform as including tools for automatic discovery and inventory of software across IT setups. (third-party source; [securityweek.com](https://www.securityweek.com/palo-alto-networks-to-acquire-koi-in-reported-400-million-transaction/), read 12 Nov 2026) - SDxCentral reports Palo Alto Networks names Claude Code, Codex CLI and OpenClaw among tools covered by the combined Cortex XDR One and AES solution. (third-party source; [sdxcentral.com](https://www.sdxcentral.com/news/palo-alto-networks-bolsters-ai-security-for-the-endpoint/), read 13 Nov 2026) - Palo Alto Networks (blog dated Oct 19, 2026) says XDR One gives a real-time inventory of AI artifacts as they appear on an endpoint. (the vendor's own claim; [paloaltonetworks.com](https://www.paloaltonetworks.com/blog/2026/10/reinventing-endpoint-security-with-cortex-xdr-one/), read 13 Nov 2026) - Palo Alto Networks says AES stops risky downloads and rogue AI agent actions before they cause damage; the action types are not named on this page. (the vendor's own claim; [paloaltonetworks.com](https://www.paloaltonetworks.com/resources/datasheets/cortex-agentic-endpoint-security-at-a-glance), read 12 Nov 2026) - The AES page promises real-time prevention: stopping malicious actions, blocking unsafe AI application interactions and preventing prompt injection. (the vendor's own claim; [paloaltonetworks.com](https://www.paloaltonetworks.com/cortex/agentic-endpoint-security.md), read 12 Nov 2026) - Cortex docs say the AES license adds runtime protection that intercepts policy-violating actions of AI coding agents before they execute (action types not named). (observed in public documentation or code; [cortex-docs.paloaltonetworks.com](https://cortex-docs.paloaltonetworks.com/cortex-xdr-6.x-rn/readme/features-introduced-in-2026/november-2026/release-highlights.md), read 13 Nov 2026) - SDxCentral reports Cortex XDR 6.0 intercepts prompts and responses at the operating system level, beyond apps with native monitoring hooks. (third-party source; [sdxcentral.com](https://www.sdxcentral.com/news/palo-alto-networks-bolsters-ai-security-for-the-endpoint/), read 13 Nov 2026) - Cortex docs say the XDR agent enforces application control that blocks untrusted applications on Windows and macOS endpoints (general application control, requires XDR ONE or AES license). (observed in public documentation or code; [cortex-docs.paloaltonetworks.com](https://cortex-docs.paloaltonetworks.com/cortex-xdr-6.x-rn/readme/features-introduced-in-2026/november-2026/feature-enhancements.md), read 13 Nov 2026) - SecurityWeek lists Koi remediation actions: quarantining risky elements, reverting versions and notifying owners (software remediation, not named agent-action blocking). (third-party source; [securityweek.com](https://www.securityweek.com/palo-alto-networks-to-acquire-koi-in-reported-400-million-transaction/), read 12 Nov 2026) - Palo Alto Networks says AES can run alongside third-party XDR products without replacing the existing stack. (the vendor's own claim; [paloaltonetworks.com](https://www.paloaltonetworks.com/resources/datasheets/cortex-agentic-endpoint-security-at-a-glance), read 12 Nov 2026) ## Checks that found nothing - A public page naming which agent actions (file read, shell command, credential access, network call) AES blocks before completion; the product pages stay at the level of risky actions and downloads: none found on 12 Nov 2026 (checked: [paloaltonetworks.com](https://www.paloaltonetworks.com/cortex/agentic-endpoint-security)) - Operating systems supported, agent footprint or time to deploy for AES, on the AES page, tech brief, datasheet or release notes: none found on 12 Nov 2026 (checked: [paloaltonetworks.com](https://www.paloaltonetworks.com/resources/datasheets/cortex-agentic-endpoint-security-at-a-glance)) - A public description of the AES audit trail, SIEM export or identity-provider integration: none found on 12 Nov 2026 (checked: [paloaltonetworks.com](https://www.paloaltonetworks.com/cortex/agentic-endpoint-security)) - A published price for AES or Koi; the page offers a personalized demo instead: none found on 12 Nov 2026 (checked: [paloaltonetworks.com](https://www.paloaltonetworks.com/cortex/agentic-endpoint-security)) - An independent analyst report (Gartner, Forrester, IDC, Software Analyst) rating Cortex Agentic Endpoint Security or Koi, in Google News results and the vendor newsroom; the only Gartner item remains the vendor-hosted note: none found on 13 Nov 2026 (checked: [news.google.com](https://news.google.com/rss/search?q=%22Koi+Security%22+customer+OR+funding+OR+Gartner+OR+Forrester+OR+analyst)) ## Position in each situation | Situation | Position | Total | Leader | | --- | --- | --- | --- | | [Coding agents](/best-for/coding-agents/) | 13th of 16 | 55.5 | Bay 75.5 | | [Regulated data](/best-for/regulated-data/) | 14th of 16 | 51.0 | Microsoft 79.0 | | [Lean team with EDR](/best-for/lean-team-edr/) | 12th of 16 | 57.5 | Bay 76.5 | | [Staff rollout](/best-for/staff-rollout/) | 14th of 16 | 48.5 | Microsoft 75.5 | | [Board in 30 days](/best-for/board-30-days/) | 12th of 16 | 59.0 | Bay 77.5 | Koi under each situation's own weights. Tied marks an equal total; the order follows the tie-break, the higher score on the heaviest criterion, then name. **About Bay.** Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). In regulated data, Bay ranks 3rd of 16 at 70.5. In staff rollout, Bay ranks 2nd of 16 at 73.5. Bay is a young company that left stealth in 2026, publishes no price and has little independent testing behind its claims. More: [all 16 platform profiles](/platforms/), [every head-to-head](/compare/), [the six agent surfaces](/surfaces/) and [the six agent actions](/actions/). ## Sources - www.paloaltonetworks.com/resources/datasheets/cortex-agentic-endpoint-security-at-a-glance: https://www.paloaltonetworks.com/resources/datasheets/cortex-agentic-endpoint-security-at-a-glance (read 12 Nov 2026) - www.paloaltonetworks.com/resources/techbriefs/koi-agentic-endpoint-security: https://www.paloaltonetworks.com/resources/techbriefs/koi-agentic-endpoint-security (read 12 Nov 2026) - www.paloaltonetworks.com/company/press/2026/palo-alto-networks-completes-acquisition-of-koi-to-secure-the-agentic-endpoint: https://www.paloaltonetworks.com/company/press/2026/palo-alto-networks-completes-acquisition-of-koi-to-secure-the-agentic-endpoint (read 12 Nov 2026) - www.securityweek.com/palo-alto-networks-to-acquire-koi-in-reported-400-million-transaction: https://www.securityweek.com/palo-alto-networks-to-acquire-koi-in-reported-400-million-transaction/ (read 12 Nov 2026) - www.sdxcentral.com/news/palo-alto-networks-bolsters-ai-security-for-the-endpoint: https://www.sdxcentral.com/news/palo-alto-networks-bolsters-ai-security-for-the-endpoint/ (read 13 Nov 2026) - www.paloaltonetworks.com/blog/2026/10/reinventing-endpoint-security-with-cortex-xdr-one: https://www.paloaltonetworks.com/blog/2026/10/reinventing-endpoint-security-with-cortex-xdr-one/ (read 13 Nov 2026) - www.paloaltonetworks.com/cortex/agentic-endpoint-security.md: https://www.paloaltonetworks.com/cortex/agentic-endpoint-security.md (read 12 Nov 2026) - cortex-docs.paloaltonetworks.com/cortex-xdr-6.x-rn/readme/features-introduced-in-2026/november-2026/release-highlights.md: https://cortex-docs.paloaltonetworks.com/cortex-xdr-6.x-rn/readme/features-introduced-in-2026/november-2026/release-highlights.md (read 13 Nov 2026) - cortex-docs.paloaltonetworks.com/cortex-xdr-6.x-rn/readme/features-introduced-in-2026/november-2026/feature-enhancements.md: https://cortex-docs.paloaltonetworks.com/cortex-xdr-6.x-rn/readme/features-introduced-in-2026/november-2026/feature-enhancements.md (read 13 Nov 2026) - www.paloaltonetworks.com/cortex/agentic-endpoint-security (checked, nothing found): https://www.paloaltonetworks.com/cortex/agentic-endpoint-security (read 12 Nov 2026) - docs-cortex.paloaltonetworks.com: https://docs-cortex.paloaltonetworks.com/ (read 12 Nov 2026) - thehackernews.com/2025/09/first-malicious-mcp-server-found.html: https://thehackernews.com/2025/09/first-malicious-mcp-server-found.html (read 12 Nov 2026) - thehackernews.com/2025/10/self-spreading-glassworm-infects-vs.html: https://thehackernews.com/2025/10/self-spreading-glassworm-infects-vs.html (read 12 Nov 2026) - www.bankinfosecurity.com/meetingtv-sues-palo-alto-networks-over-koi-threat-report-a-32136: https://www.bankinfosecurity.com/meetingtv-sues-palo-alto-networks-over-koi-threat-report-a-32136 (read 13 Nov 2026) - www.sdxcentral.com/news/palo-alto-networks-sued-over-ai-slop-with-questions-over-400m-koi-acquisition-report: https://www.sdxcentral.com/news/palo-alto-networks-sued-over-ai-slop-with-questions-over-400m-koi-acquisition-report/ (read 13 Nov 2026) - www.calcalistech.com/ctechnews/article/xpz78znuc: https://www.calcalistech.com/ctechnews/article/xpz78znuc (read 12 Nov 2026) - news.google.com/rss/search?q=%22Koi+Security%22+customer+OR+funding+OR+Gartner+OR+Forrester+OR+analyst (checked, nothing found): https://news.google.com/rss/search?q=%22Koi+Security%22+customer+OR+funding+OR+Gartner+OR+Forrester+OR+analyst (read 13 Nov 2026) - www.paloaltonetworks.com/blog/2026/02/securing-the-agentic-endpoint: https://www.paloaltonetworks.com/blog/2026/02/securing-the-agentic-endpoint/ (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/platforms/koi/ --- # Lasso Security: 13th of 16 at 53.0, strongest on Proof (named customers, funding, analyst coverage) at 8 of 10 PLATFORM PROFILE, read 13 Nov 2026 **The short answer.** Lasso Security ranks 13th of 16 AI endpoint security tools at 53.0 of 100. Lasso Security and Pillar Security tie at 53.0 and both score 6 on Sees, the heaviest criterion here, so name order puts Lasso Security 13th. Lasso Security shows 3 of 6 agent surfaces in full and can stop 1 of 6 agent actions, as documented. ## Profile - **Owner:** Lasso Security Ltd. (independent; no parent or acquirer named) - **Product:** Lasso AI Security Platform (AI Usage Control, AI Agents Security, Intent Security) - **Status:** Active under this name, checked 13 Nov 2026 - **Category:** AI security platform with usage control, agent runtime protection and red teaming - **Deployment:** Hook-based integration via enterprise management platform (no endpoint agent stated for coding assistants); API/gateway for applications; Lasso also says it connects to an existing EDR or MDM - **Operating systems:** none found on 12 Nov 2026 - **Time to deploy:** "Getting started takes minutes" (vendor claim, coding assistants page) - **Price:** none found on 12 Nov 2026 (pricing page returns not found; Book a Demo only) - **Website:** [lasso.security](https://www.lasso.security/) Scores out of 10 (weight)Sees 256Stops 255Deployment 155Beyond 105Policy 105Transparency 104Proof 58 ## Strength and weakness **Strength:** Lasso documents inspecting every coding-agent tool call before execution via Claude Code, Cursor and Codex hooks, with out-of-scope calls flagged or blocked, plus an open-source MCP gateway. **Weakness:** None found on 12 Nov 2026: a public docs site (docs.lasso.security is login-gated), a published price, or an OS list; the open-source Claude hook only warns. ## Scores: 53.0 of 100, 13th of 16 | Criterion | Weight | Score | Why this score | | --- | --- | --- | --- | | Sees agentic activity on the endpoint | 25 | 6 | The vendor inventories MCP connections for named desktop and coding agents, and an observed open-source scanner checks MCP servers. No browser extension inventory was found on 12 Nov 2026. | | Stops the action | 25 | 5 | The vendor says it blocks tool calls outside the allowed scope and risky MCP servers, and enforces some agent policies at the proxy or gateway layer; observed code shows gateway blocking and a warn-only hook. Named blocking of file reads, package installs or credential access was none found on 12 Nov 2026. | | Deployment friction | 15 | 5 | The vendor says it is agentless through the enterprise management platform and takes minutes to set up. An operating-system list and an independent deployment source were none found on 13 Nov 2026. | | Beyond the endpoint | 10 | 5 | The vendor says one policy set covers SaaS chatbots, desktop agents and MCP, finds agents across SaaS, cloud and builder platforms, and reads a SaaS agent platform through an API. Browser extension coverage was none found on 12 Nov 2026, and an independent source was none found on 13 Nov 2026. | | Policy and governance | 10 | 5 | The vendor offers audit trails, SIEM export of MCP logs and framework templates. A named EDR or identity-provider integration was none found on 12 Nov 2026, and an independent source was none found on 13 Nov 2026. | | Transparency | 10 | 4 | The vendor publishes research and open-source tools, confirmed by third-party and observed sources. Docs are login-gated and no published price was found on 12 Nov 2026. | | Proof | 5 | 8 | Third-party press confirms $30 million raised and names customers including BMW and a U.S. federal department, and reports a regional reseller; the vendor cites Gartner and Latio placements. The analyst placements are vendor-stated. | Lasso Security: weighted total 53.0 of 100 on the published weights. ## What does Lasso Security see on an endpoint? Lasso Security documents 3 of 6 agent surfaces in full (IDE assistants, command-line agents and MCP servers) and 2 in part (desktop apps and SaaS copilots). Nothing was found for browser extensions. Each cell with its note and source: [Lasso Security on the Agent Surface Map](/agent-surface-map/lasso-security/). ## Which agent actions can Lasso Security stop? Lasso Security documents stopping 1 of 6 agent actions before they complete (data paste) and 3 in part (shell command, network call and credential access). Nothing was found for file read and package install. Each cell with its note and source: [Lasso Security on the Action Control Matrix](/action-control-matrix/lasso-security/). ## How much does Lasso Security cost? Lasso Security publishes no list price. None found on 12 Nov 2026 (pricing page returns not found; Book a Demo only) Price counts toward no score; deployment friction is scored under Deployment. ## Evidence behind the scores - Lasso says it discovers and inventories AI assets from models and SaaS chatbots to local desktop agents and homegrown agents with their connected tools. (the vendor's own claim; [lasso.security](https://www.lasso.security/platform/ai-usage-control), read 12 Nov 2026) - Lasso's MCP page says it inventories Claude Code, Claude Desktop, Cursor, Windsurf and custom agent MCP connections with a risk score per server. (the vendor's own claim; [lasso.security](https://www.lasso.security/use-cases/mcp-security), read 12 Nov 2026) - Lasso says it secures Claude Code, Cursor, Codex and OpenCode coding assistants without separate deployments for each. (the vendor's own claim; [lasso.security](https://www.lasso.security/use-cases/ai-coding-assistants), read 12 Nov 2026) - Lasso says its endpoint discovery profiles every AI tool, coding agent and MCP or skill connection to build the inventory policy is written against. (the vendor's own claim; [lasso.security](https://www.lasso.security/blog/securing-ai-on-endpoints-how-to-govern-ai-tools-coding-agents-and-mcp-connections), read 12 Nov 2026) - Lasso's open-source MCP Gateway repository (403 stars on 12 Nov 2026) includes a security scanner that checks MCP server reputation before loading. (observed in public documentation or code; [github.com](https://github.com/lasso-security/mcp-gateway), read 12 Nov 2026) - Lasso says it inspects every coding-agent tool call before execution and flags or blocks anything outside the permitted tool scope. (the vendor's own claim; [lasso.security](https://www.lasso.security/use-cases/ai-coding-assistants), read 12 Nov 2026) - Lasso says MCP servers with a high risk score can be managed or blocked, and that injection attempts are detected and blocked at the connection layer. (the vendor's own claim; [lasso.security](https://www.lasso.security/use-cases/mcp-security), read 12 Nov 2026) - Lasso says MCP data loss prevention detects API keys and credentials in tool calls and masks them before they reach external servers. (the vendor's own claim; [lasso.security](https://www.lasso.security/use-cases/mcp-security), read 12 Nov 2026) - Lasso's open-source Kong plugin classifies AI chat traffic and blocks or masks it inline in the request path, evidence of gateway-level enforcement. (observed in public documentation or code; [github.com](https://github.com/lasso-security/kong-ai-lasso-guardrail), read 12 Nov 2026) - Lasso's open-source claude-hooks repository ships a prompt injection defender that scans tool outputs and warns Claude, which is a warn action rather than a block. (observed in public documentation or code; [github.com](https://github.com/lasso-security/claude-hooks), read 12 Nov 2026) - Lasso's rogue-agents guide says policies are enforced inline at the proxy, API or AI Gateway layer (a gateway enforcement point for agents it protects, not an endpoint hook); this limits where the stated block applies and does not contradict lasso-security-07. (the vendor's own claim; [lasso.security](https://www.lasso.security/blog/rogue-ai-agents-what-they-are-how-they-happen-and-how-to-stop-them), read 13 Nov 2026) - Lasso says it deploys through the enterprise management platform with no agent to install and no code to change. (the vendor's own claim; [lasso.security](https://www.lasso.security/use-cases/ai-coding-assistants), read 12 Nov 2026) - Lasso says getting started takes minutes via Claude Code lifecycle hooks. (the vendor's own claim; [lasso.security](https://www.lasso.security/use-cases/ai-coding-assistants), read 12 Nov 2026) - Lasso says its usage control spans SaaS and web chatbots, local desktop agents and MCPs in one set of policies. (the vendor's own claim; [lasso.security](https://www.lasso.security/platform/ai-usage-control), read 12 Nov 2026) ## Checks that found nothing - Lasso pages naming blocking of file reads, package installs or credential-file access by desktop or CLI agents as distinct action types: none found on 12 Nov 2026 (checked: [lasso.security](https://www.lasso.security/use-cases/ai-coding-assistants)) - An operating-system support list for any Lasso endpoint component on its product pages: none found on 12 Nov 2026 (checked: [lasso.security](https://www.lasso.security/use-cases/ai-coding-assistants)) - A third-party or observed source (trade press, marketplace listing, public docs) stating Lasso's deployment model, operating systems or time to deploy; Google News searches on agentless, deploy, macOS and Windows returned only vendor-sourced or unrelated items: none found on 13 Nov 2026 (checked: [news.google.com](https://news.google.com/rss/search?q=%22Lasso+Security%22+agentless+OR+deploy+OR+macOS+OR+Windows)) - A documented browser extension or browser-agent inventory in Lasso's usage control page: none found on 12 Nov 2026 (checked: [lasso.security](https://www.lasso.security/platform/ai-usage-control)) - A third-party or observed source documenting Lasso coverage of browsers or SaaS copilots in one policy; Google News searches on browser, SaaS and copilot returned no Lasso product coverage: none found on 13 Nov 2026 (checked: [news.google.com](https://news.google.com/rss/search?q=%22Lasso+Security%22+browser+OR+SaaS+OR+copilot)) - Documented EDR or identity-provider integrations (beyond role-based permissions) on Lasso's public product pages: none found on 12 Nov 2026 (checked: [lasso.security](https://www.lasso.security/platform/ai-usage-control)) - A third-party or observed source documenting Lasso SIEM, Splunk, Okta or audit-trail integrations; Google News searches returned list articles without integration detail and docs.lasso.security remained login-gated: none found on 13 Nov 2026 (checked: [news.google.com](https://news.google.com/rss/search?q=%22Lasso+Security%22+SIEM+OR+Splunk+OR+Okta+OR+audit)) - A published price; lasso.security/pricing returns a not-found page and product pages route to Book a Demo: none found on 12 Nov 2026 (checked: [lasso.security](https://www.lasso.security/pricing)) ## Position in each situation | Situation | Position | Total | Leader | | --- | --- | --- | --- | | [Coding agents](/best-for/coding-agents/) | 14th of 16 | 53.5 | Bay 75.5 | | [Regulated data](/best-for/regulated-data/) | 12th of 16 | 54.0 | Microsoft 79.0 | | [Lean team with EDR](/best-for/lean-team-edr/) | 14th of 16 | 54.0 | Bay 76.5 | | [Staff rollout](/best-for/staff-rollout/) | 12th of 16 | 53.0 | Microsoft 75.5 | | [Board in 30 days](/best-for/board-30-days/) | 13th of 16 | 55.0 | Bay 77.5 | Lasso Security under each situation's own weights. Tied marks an equal total; the order follows the tie-break, the higher score on the heaviest criterion, then name. ## Head-to-heads - [Knostic vs Lasso Security](/compare/knostic-vs-lasso-security/): Knostic 64.0 to Lasso Security 53.0 on the published weights **About Bay.** Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). In regulated data, Bay ranks 3rd of 16 at 70.5. In staff rollout, Bay ranks 2nd of 16 at 73.5. Bay is a young company that left stealth in 2026, publishes no price and has little independent testing behind its claims. More: [all 16 platform profiles](/platforms/), [every head-to-head](/compare/), [the six agent surfaces](/surfaces/) and [the six agent actions](/actions/). ## Sources - www.lasso.security/platform/ai-usage-control: https://www.lasso.security/platform/ai-usage-control (read 12 Nov 2026) - www.lasso.security/use-cases/mcp-security: https://www.lasso.security/use-cases/mcp-security (read 12 Nov 2026) - www.lasso.security/use-cases/ai-coding-assistants: https://www.lasso.security/use-cases/ai-coding-assistants (read 12 Nov 2026) - www.lasso.security/blog/securing-ai-on-endpoints-how-to-govern-ai-tools-coding-agents-and-mcp-connections: https://www.lasso.security/blog/securing-ai-on-endpoints-how-to-govern-ai-tools-coding-agents-and-mcp-connections (read 12 Nov 2026) - github.com/lasso-security/mcp-gateway: https://github.com/lasso-security/mcp-gateway (read 12 Nov 2026) - github.com/lasso-security/kong-ai-lasso-guardrail: https://github.com/lasso-security/kong-ai-lasso-guardrail (read 12 Nov 2026) - github.com/lasso-security/claude-hooks: https://github.com/lasso-security/claude-hooks (read 12 Nov 2026) - www.lasso.security/blog/rogue-ai-agents-what-they-are-how-they-happen-and-how-to-stop-them: https://www.lasso.security/blog/rogue-ai-agents-what-they-are-how-they-happen-and-how-to-stop-them (read 13 Nov 2026) - news.google.com/rss/search?q=%22Lasso+Security%22+agentless+OR+deploy+OR+macOS+OR+Windows (checked, nothing found): https://news.google.com/rss/search?q=%22Lasso+Security%22+agentless+OR+deploy+OR+macOS+OR+Windows (read 13 Nov 2026) - www.lasso.security: https://www.lasso.security (read 12 Nov 2026) - www.lasso.security/blog/bringing-claude-enterprise-activity-into-lasso-with-claudes-compliance-api: https://www.lasso.security/blog/bringing-claude-enterprise-activity-into-lasso-with-claudes-compliance-api (read 12 Nov 2026) - news.google.com/rss/search?q=%22Lasso+Security%22+browser+OR+SaaS+OR+copilot (checked, nothing found): https://news.google.com/rss/search?q=%22Lasso+Security%22+browser+OR+SaaS+OR+copilot (read 13 Nov 2026) - news.google.com/rss/search?q=%22Lasso+Security%22+SIEM+OR+Splunk+OR+Okta+OR+audit (checked, nothing found): https://news.google.com/rss/search?q=%22Lasso+Security%22+SIEM+OR+Splunk+OR+Okta+OR+audit (read 13 Nov 2026) - www.theregister.com/ai-and-ml/2026/09/17/ai-model-watermarking-changes-agent-behavior/5296998: https://www.theregister.com/ai-and-ml/2026/09/17/ai-model-watermarking-changes-agent-behavior/5296998 (read 12 Nov 2026) - docs.lasso.security: https://docs.lasso.security/ (read 12 Nov 2026) - www.lasso.security/pricing (checked, nothing found): https://www.lasso.security/pricing (read 12 Nov 2026) - siliconangle.com/2026/09/02/lasso-security-launches-cpu-only-ai-guardrail-raises-30m: https://siliconangle.com/2026/09/02/lasso-security-launches-cpu-only-ai-guardrail-raises-30m/ (read 12 Nov 2026) - finance.yahoo.com/technology/ai/articles/lasso-security-announces-future-ai-080000059.html: https://finance.yahoo.com/technology/ai/articles/lasso-security-announces-future-ai-080000059.html (read 12 Nov 2026) - techcentral.co.za/neworder-lasso-security-ai-agent-controls/285558: https://techcentral.co.za/neworder-lasso-security-ai-agent-controls/285558/ (read 13 Nov 2026) - thehackernews.com/2026/04/google-fixes-cvss-10-gemini-cli-ci-rce.html: https://thehackernews.com/2026/04/google-fixes-cvss-10-gemini-cli-ci-rce.html (read 12 Nov 2026) - thehackernews.com/2026/10/opencode-flaw-lets-malicious-mcp.html: https://thehackernews.com/2026/10/opencode-flaw-lets-malicious-mcp.html (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/platforms/lasso-security/ --- # Pillar Security: 14th of 16 at 53.0, strongest on Deployment at 6 of 10 PLATFORM PROFILE, read 13 Nov 2026 **The short answer.** Pillar Security ranks 14th of 16 AI endpoint security tools at 53.0 of 100. Lasso Security and Pillar Security tie at 53.0 and both score 6 on Sees, the heaviest criterion here, so name order puts Lasso Security 13th. Pillar Security stands out on Deployment at 6 of 10, and its lowest score is Beyond at 2 of 10. ## Profile - **Owner:** Pillar Security; Vanta signed a definitive agreement to acquire it on Nov 9, 2026 (announced, closing not confirmed on pages read) - **Product:** Pillar Agentic Endpoint (Pillar for AI Coding Agents, Runtime Guardian) within the Pillar AI security platform - **Status:** Active under this name, checked 13 Nov 2026 - **Category:** AI security platform with an endpoint module for coding agents - **Deployment:** Endpoint scripts pushed through existing MDM (Jamf Pro, Kandji, JumpCloud), with a lightweight per-user service and hooks written into supported agents - **Operating systems:** macOS, Linux, Windows - **Time to deploy:** none found on 12 Nov 2026 - **Price:** none found on 12 Nov 2026 (pillar.security/pricing returned 404; demo-only) - **Website:** [pillar.security](https://www.pillar.security) Scores out of 10 (weight)Sees 256Stops 256Deployment 156Beyond 102Policy 105Transparency 104Proof 56 ## Strength and weakness **Strength:** Documents pre-action blocking in the agent's own pre-tool hook for Claude Code, Codex CLI and Cursor, with credential-exfiltration and persistence controls and MDM-installed hooks developers cannot disable. **Weakness:** Coverage is coding agents only: none found on 12 Nov 2026 for desktop AI apps, browsers or SaaS copilots, pre-action enforcement lists three agents, and an acquisition by Vanta is pending. ## Scores: 53.0 of 100, 14th of 16 | Criterion | Weight | Score | Why this score | | --- | --- | --- | --- | | Sees agentic activity on the endpoint | 25 | 6 | The vendor names coding agents and inventories MCP servers, skills, plugins and hooks, with posture checks; it says it detects 18 coding agents but parses the configuration of only five. Desktop app or browser coverage and a second source were none found on 13 Nov 2026. | | Stops the action | 25 | 6 | The vendor says a pre-tool hook blocks credential reads followed by egress, public uploads, persistence writes, cloud metadata requests and unlisted MCP tools before they run. This is limited to three agents, the hooks fail open, and all of it is vendor claim. | | Deployment friction | 15 | 6 | The vendor deploys light scripts via MDM on macOS, Linux and Windows. A time-to-deploy figure and an independent deployment source were none found on 13 Nov 2026. | | Beyond the endpoint | 10 | 2 | The wider platform covers homegrown AI and gateways. No browser, cloud workspace or SaaS-agent coverage under the endpoint policy was found on 12 Nov 2026. | | Policy and governance | 10 | 5 | The vendor offers Off, Monitor and Block per control, a session timeline and session buffer for decisions, and a data API with Splunk and Jira integration. A named EDR or identity-provider integration was none found on 13 Nov 2026. | | Transparency | 10 | 4 | The vendor publishes agent-vulnerability research that third-party press credits, including a new MCP Inspector flaw. A published price and public documentation (access-code gate observed) were none found on 12 Nov 2026. | | Proof | 5 | 6 | Third-party press confirms the funding and an acquisition at about $150 million, and the vendor quotes named customers and cites Gartner recognition. The customer names and analyst mention are vendor-published. | Pillar Security: weighted total 53.0 of 100 on the published weights. ## What does Pillar Security see on an endpoint? Pillar Security documents 3 of 6 agent surfaces in full (IDE assistants, command-line agents and MCP servers) and 0 in part. Nothing was found for desktop apps, browser extensions and SaaS copilots. Each cell with its note and source: [Pillar Security on the Agent Surface Map](/agent-surface-map/pillar-security/). ## Which agent actions can Pillar Security stop? Pillar Security documents stopping 4 of 6 agent actions before they complete (file read, shell command, network call and credential access) and 0 in part. Nothing was found for data paste and package install. Each cell with its note and source: [Pillar Security on the Action Control Matrix](/action-control-matrix/pillar-security/). ## How much does Pillar Security cost? Pillar Security publishes no list price. None found on 12 Nov 2026 (pillar.security/pricing returned 404; demo-only) Price counts toward no score; deployment friction is scored under Deployment. ## Evidence behind the scores - Pillar's Agentic Endpoint page says it inventories coding agents, MCP servers, skills, plugins, subagents and hooks on every machine, deployed through MDM. (the vendor's own claim; [pillar.security](https://www.pillar.security/platform/agentic-endpoint), read 12 Nov 2026) - Pillar says it runs deep scans for Claude Code, Cursor, Codex CLI, Gemini CLI, OpenCode and Pi, and fingerprints Amazon Q, Antigravity, OpenClaw and others. (the vendor's own claim; [pillar.security](https://www.pillar.security/platform/agentic-endpoint), read 12 Nov 2026) - Pillar says it scores local and remote MCP servers against a catalog of 23,000+ pre-analyzed servers and shows which agents and endpoints load each one. (the vendor's own claim; [pillar.security](https://www.pillar.security/platform/agentic-endpoint), read 12 Nov 2026) - Pillar says it flags YOLO mode, Cursor auto-run, Codex approval bypass, disabled workspace trust and missing sandboxes; posture scans read agent configuration, not source code. (the vendor's own claim; [pillar.security](https://www.pillar.security/platform/agentic-endpoint), read 12 Nov 2026) - Pillar's guide says its scanner detects 18 coding agents but parses the configuration of only five. (the vendor's own claim; [pillar.security](https://www.pillar.security/blog/best-ai-coding-agent-security-tools-for-the-enterprise-2026), read 13 Nov 2026) - Pillar's guide says its OpenCode configuration coverage is narrower than for the other parsed agents (a limit on depth). (the vendor's own claim; [pillar.security](https://www.pillar.security/blog/best-ai-coding-agent-security-tools-for-the-enterprise-2026), read 13 Nov 2026) - Pillar says Runtime Guardian sits in the agent's own pre-tool hook for Claude Code, Codex CLI and Cursor and decides on every action before it runs. (the vendor's own claim; [pillar.security](https://www.pillar.security/platform/agentic-endpoint), read 12 Nov 2026) - Pillar says it blocks reads of keys.env, SSH and cloud credentials from reaching external destinations. (the vendor's own claim; [pillar.security](https://www.pillar.security/platform/agentic-endpoint), read 12 Nov 2026) - Pillar says it blocks committed git hooks that write persistence and agents rewriting their own configuration, and offers Off, Monitor or Block per control with confirmation prompts. (the vendor's own claim; [pillar.security](https://www.pillar.security/platform/agentic-endpoint), read 12 Nov 2026) - Pillar says per-server MCP tool allow-lists restrict which tools agents may call, in Monitor or Block mode. (the vendor's own claim; [pillar.security](https://www.pillar.security/platform/agentic-endpoint), read 12 Nov 2026) - Pillar's guide says five controls ship in Block mode by default (public uploads, credential reads followed by egress, git-hook persistence, and others) and cloud metadata requests are always blocked. (the vendor's own claim; [pillar.security](https://www.pillar.security/blog/best-ai-coding-agent-security-tools-for-the-enterprise-2026), read 13 Nov 2026) - Pillar says pre-action enforcement is for Claude Code, Codex CLI and Cursor only (as of October 2026); Gemini CLI, OpenCode and Pi are scanned for configuration but runtime blocking was not listed for them. (the vendor's own claim; [pillar.security](https://www.pillar.security/platform/agentic-endpoint), read 12 Nov 2026) - Pillar's own October 2026 buyer's guide says hook-based enforcement, Pillar's included per its documentation, fails open, so a broken or unreachable hook does not stop the agent (a limit on pre-action blocking). (the vendor's own claim; [pillar.security](https://www.pillar.security/blog/best-ai-coding-agent-security-tools-for-the-enterprise-2026), read 13 Nov 2026) - Pillar says it deploys through existing MDM as a script on macOS, Linux and Windows (Jamf Pro, Kandji, JumpCloud or any MDM that runs scripts), with a lightweight per-user service for Runtime Guardian. (the vendor's own claim; [pillar.security](https://www.pillar.security/platform/agentic-endpoint), read 12 Nov 2026) ## Checks that found nothing - An independent (third-party or observed) source for Pillar's agent inventory and MCP catalog claims; Google News, Bing News, GitHub and NVD returned only Pillar's own pages and acquisition coverage: none found on 13 Nov 2026 (checked: [news.google.com](https://news.google.com/rss/search?q=%22Pillar+Security%22&hl=en-US&gl=US&ceid=US:en)) - A published time-to-deploy figure on Pillar's endpoint page, coding agents launch post or buyer's guide: none found on 12 Nov 2026 (checked: [pillar.security](https://www.pillar.security/platform/agentic-endpoint)) - An independent source for Pillar's MDM deployment and operating-system coverage; only Pillar's own pages describe it: none found on 13 Nov 2026 (checked: [news.google.com](https://news.google.com/rss/search?q=%22Pillar+Security%22&hl=en-US&gl=US&ceid=US:en)) - Pillar documentation of browser extensions or browser agents on employee browsers, or Microsoft 365 Copilot and Salesforce agent coverage, on its endpoint page: none found on 12 Nov 2026 (checked: [pillar.security](https://www.pillar.security/platform/agentic-endpoint)) - A published price (pillar.security/pricing returned a 404 page) or a public documentation site; the site has a blog, SAIL framework and research pages: none found on 12 Nov 2026 (checked: [pillar.security](https://www.pillar.security/pricing)) ## Position in each situation | Situation | Position | Total | Leader | | --- | --- | --- | --- | | [Coding agents](/best-for/coding-agents/) | 12th of 16 | 56.0 | Bay 75.5 | | [Regulated data](/best-for/regulated-data/) | 15th of 16 | 50.5 | Microsoft 79.0 | | [Lean team with EDR](/best-for/lean-team-edr/) | 13th of 16 | 56.0 | Bay 76.5 | | [Staff rollout](/best-for/staff-rollout/) | 16th of 16 (tied) | 48.0 | Microsoft 75.5 | | [Board in 30 days](/best-for/board-30-days/) | 14th of 16 | 54.5 | Bay 77.5 | Pillar Security under each situation's own weights. Tied marks an equal total; the order follows the tie-break, the higher score on the heaviest criterion, then name. ## Head-to-heads - [Bay vs Pillar Security](/compare/bay-vs-pillar-security/): Bay 71.5 to Pillar Security 53.0 on the published weights - [Pillar Security vs Zenity](/compare/pillar-security-vs-zenity/): Zenity 65.0 to Pillar Security 53.0 on the published weights **About Bay.** Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). In regulated data, Bay ranks 3rd of 16 at 70.5. In staff rollout, Bay ranks 2nd of 16 at 73.5. Bay is a young company that left stealth in 2026, publishes no price and has little independent testing behind its claims. More: [all 16 platform profiles](/platforms/), [every head-to-head](/compare/), [the six agent surfaces](/surfaces/) and [the six agent actions](/actions/). ## Sources - www.pillar.security/platform/agentic-endpoint: https://www.pillar.security/platform/agentic-endpoint (read 12 Nov 2026) - www.pillar.security/blog/best-ai-coding-agent-security-tools-for-the-enterprise-2026: https://www.pillar.security/blog/best-ai-coding-agent-security-tools-for-the-enterprise-2026 (read 13 Nov 2026) - news.google.com/rss/search?q=%22Pillar+Security%22&hl=en-US&gl=US&ceid=US:en (checked, nothing found): https://news.google.com/rss/search?q=%22Pillar+Security%22&hl=en-US&gl=US&ceid=US:en (read 13 Nov 2026) - www.pillar.security/blog/introducing-pillar-for-ai-coding-agents: https://www.pillar.security/blog/introducing-pillar-for-ai-coding-agents (read 12 Nov 2026) - www.pillar.security/blog/how-to-evaluate-endpoint-security-for-ai-agents: https://www.pillar.security/blog/how-to-evaluate-endpoint-security-for-ai-agents (read 12 Nov 2026) - thehackernews.com/2026/04/google-fixes-cvss-10-gemini-cli-ci-rce.html: https://thehackernews.com/2026/04/google-fixes-cvss-10-gemini-cli-ci-rce.html (read 12 Nov 2026) - thehackernews.com/2026/10/opencode-flaw-lets-malicious-mcp.html: https://thehackernews.com/2026/10/opencode-flaw-lets-malicious-mcp.html (read 12 Nov 2026) - www.infosecurity-magazine.com/news/two-critical-flaws-in-n8n-ai: https://www.infosecurity-magazine.com/news/two-critical-flaws-in-n8n-ai/ (read 12 Nov 2026) - www.pillar.security/blog/inspecting-the-inspector-rce-in-mcps-official-debugging-tool: https://www.pillar.security/blog/inspecting-the-inspector-rce-in-mcps-official-debugging-tool (read 13 Nov 2026) - www.pillar.security/pricing (checked, nothing found): https://www.pillar.security/pricing (read 12 Nov 2026) - docs.pillar.security: https://docs.pillar.security (read 12 Nov 2026) - www.calcalistech.com/ctechnews/article/hkjzjkjrfl: https://www.calcalistech.com/ctechnews/article/hkjzjkjrfl (read 12 Nov 2026) - www.pillar.security/about: https://www.pillar.security/about (read 12 Nov 2026) - pulse2.com/vanta-to-acquire-pillar-security-to-expand-ai-agent-protection-as-customer-base-surpasses-20000: https://pulse2.com/vanta-to-acquire-pillar-security-to-expand-ai-agent-protection-as-customer-base-surpasses-20000/ (read 12 Nov 2026) - github.com/lasso-security/mcp-gateway: https://github.com/lasso-security/mcp-gateway (read 12 Nov 2026) - github.com/lasso-security/kong-ai-lasso-guardrail: https://github.com/lasso-security/kong-ai-lasso-guardrail (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/platforms/pillar-security/ --- # LayerX: 15th of 16 at 50.0, strongest on Policy at 6 of 10 PLATFORM PROFILE, read 13 Nov 2026 **The short answer.** LayerX ranks 15th of 16 AI endpoint security tools at 50.0 of 100. LayerX stands out on Policy at 6 of 10, and its lowest score is Transparency at 4 of 10. LayerX shows 2 of 6 agent surfaces in full and can stop 1 of 6 agent actions, as documented. ## Profile - **Listed as:** LayerX (now Akamai Workforce Protector) - **Owner:** Akamai Technologies (acquired LayerX; announced May 14, 2026) - **Product:** Akamai Workforce Protector (formerly LayerX): browser extension, management console and cloud intelligence for AI usage control - **Status:** Active under this name, checked 13 Nov 2026 - **Category:** Browser-extension AI usage control and browser security, now part of Akamai Zero Trust - **Deployment:** Browser extension (Chrome, Edge, Firefox, Safari) with cloud management console and desktop visibility component; no proxy, as published - **Operating systems:** none found on 12 Nov 2026 - **Time to deploy:** none found on 12 Nov 2026 (published as "deploy quickly" without a figure) - **Price:** none found on 13 Nov 2026 (Akamai page offers a demo form only; listed on AWS Marketplace with pay-as-you-go pricing and no figure in the release) - **Website:** [layerxsecurity.com](https://layerxsecurity.com) Scores out of 10 (weight)Sees 255Stops 255Deployment 155Beyond 105Policy 106Transparency 104Proof 55 ## Strength and weakness **Strength:** It enforces warn, redact, block or guide policy on prompts, uploads and copy-paste at the moment of interaction with no proxy or browser replacement, and covers agentic browsers. **Weakness:** None found on 12 Nov 2026: documented control over IDE assistants, command-line coding agents or local MCP servers, or blocking of shell, file or network actions outside the browser; no public price or open docs. ## Scores: 50.0 of 100, 15th of 16 | Criterion | Weight | Score | Why this score | | --- | --- | --- | --- | | Sees agentic activity on the endpoint | 25 | 5 | Third-party review confirms browser-session inventory of apps and extensions, the vendor names agentic browsers, and it says a native desktop agent covers desktop AI apps such as ChatGPT, Claude and Copilot desktop. Inventory of IDE assistants, command-line agents or MCP servers was none found on 12 Nov 2026. | | Stops the action | 25 | 5 | The vendor blocks prompts, pastes, uploads and risky extensions in the browser and says the same DLP applies on desktop, with the browser part confirmed by a third-party review. Blocking of agent shell commands, file reads, network calls or credential access was none found on 12 Nov 2026. | | Deployment friction | 15 | 5 | Deployment is a lightweight browser extension with no proxy, pushed by group policy or MDM per a third-party review, plus a desktop agent installed through standard tooling. An operating-system list and a time-to-deploy figure were none found on 13 Nov 2026. | | Beyond the endpoint | 10 | 5 | The vendor covers browsers, SaaS apps and agentic browsers under its browser policy. Cloud workspace or SaaS-hosted agent coverage, and an independent confirmation, were none found on 13 Nov 2026. | | Policy and governance | 10 | 6 | A central console, IdP and SIEM integrations, a cloud security console feed and per-identity attribution are listed, and a third-party review describes the policy model. An EDR integration was none found on 12 Nov 2026. | | Transparency | 10 | 4 | The vendor publishes agent and browser research that third-party press reports, and it sells through a cloud marketplace. A published price figure and public documentation (docs redirect to sign-in) were none found on 12 Nov 2026. | | Proof | 5 | 5 | Third-party press confirms seed and Series A funding, and the vendor reports an acquisition. No named customers or analyst ranking were found on 12 Nov 2026. | LayerX: weighted total 50.0 of 100 on the published weights. ## What does LayerX see on an endpoint? LayerX documents 2 of 6 agent surfaces in full (desktop apps and browser extensions) and 1 in part (SaaS copilots). Nothing was found for IDE assistants, command-line agents and MCP servers. Each cell with its note and source: [LayerX on the Agent Surface Map](/agent-surface-map/layerx/). ## Which agent actions can LayerX stop? LayerX documents stopping 1 of 6 agent actions before they complete (data paste) and 1 in part (package install). Nothing was found for file read, shell command, network call and credential access. Each cell with its note and source: [LayerX on the Action Control Matrix](/action-control-matrix/layerx/). ## How much does LayerX cost? LayerX publishes no list price. None found on 13 Nov 2026 (Akamai page offers a demo form only; listed on AWS Marketplace with pay-as-you-go pricing and no figure in the release) Price counts toward no score; deployment friction is scored under Deployment. ## Evidence behind the scores - Akamai says the extension monitors and enforces policy on user and AI agent actions in the browser, naming ChatGPT Atlas, Perplexity Comet and Dia. (the vendor's own claim; [akamai.com](https://www.akamai.com/products/akamai-workforce-protector), read 12 Nov 2026) - Akamai says Workforce Protector analyzes browser extension risk scores, behaviors and permissions with a risk database, and blocks malicious or overprivileged extensions. (the vendor's own claim; [akamai.com](https://www.akamai.com/products/akamai-workforce-protector), read 12 Nov 2026) - The Hacker News review describes a Discovery page that inventories accounts, apps and browser extensions seen in browsing sessions, with risky extensions flagged and blockable. (third-party source; [thehackernews.com](https://thehackernews.com/2023/11/hands-on-review-layerxs-enterprise.html), read 12 Nov 2026) - LayerX's February 2026 agentic-browser blog says it identifies AI browsers explicitly and shows which agentic browsers are installed, by user and device. (the vendor's own claim; [akamai.com](https://www.akamai.com/blog/security/layerx-announces-the-first-dedicated-solution-for-agentic-browser-protection), read 12 Nov 2026) - LayerX says (GlobeNewswire release of May 20, 2026, carried by the Manila Times) that it enforces AI usage policy on native desktop AI apps such as ChatGPT desktop, Claude desktop and Microsoft 365 Copilot, using a native desktop agent. (the vendor's own claim; [manilatimes.net](https://www.manilatimes.net/2026/05/21/tmt-newswire/globenewswire/layerx-security-integrates-with-new-extended-plan-for-aws-security-hub/2348650), read 13 Nov 2026) - Akamai says policies can warn, redact, block or guide user actions based on context and risk. (the vendor's own claim; [akamai.com](https://www.akamai.com/products/akamai-workforce-protector), read 12 Nov 2026) - Akamai lists DLP over text input, copy and paste, uploads, downloads and data exchanges. (the vendor's own claim; [akamai.com](https://www.akamai.com/products/akamai-workforce-protector), read 12 Nov 2026) - The Hacker News review describes policy actions that include blocking access, preventing uploads, disabling risky page components and warning the user. (third-party source; [thehackernews.com](https://thehackernews.com/2023/11/hands-on-review-layerxs-enterprise.html), read 12 Nov 2026) - LayerX says it can disable or restrict the AI side panel in agentic browsers when users reach sensitive enterprise assets. (the vendor's own claim; [akamai.com](https://www.akamai.com/blog/security/layerx-announces-the-first-dedicated-solution-for-agentic-browser-protection), read 12 Nov 2026) - LayerX says it applies DLP to prompts, file uploads and copy/paste before data reaches an AI model, on both browser and desktop surfaces. (the vendor's own claim; [manilatimes.net](https://www.manilatimes.net/2026/05/21/tmt-newswire/globenewswire/layerx-security-integrates-with-new-extended-plan-for-aws-security-hub/2348650), read 13 Nov 2026) - Akamai says Workforce Protector deploys without proxies, network redesign or browser replacement, as a lightweight extension. (the vendor's own claim; [akamai.com](https://www.akamai.com/products/akamai-workforce-protector), read 12 Nov 2026) - Akamai says it applies controls across Chrome, Edge, Firefox and Safari. (the vendor's own claim; [akamai.com](https://www.akamai.com/products/akamai-workforce-protector), read 12 Nov 2026) - The Hacker News review says LayerX deploys through group policy, any enterprise device management platform or the browser's management workspace; in the last case nothing is installed on the device. (third-party source; [thehackernews.com](https://thehackernews.com/2023/11/hands-on-review-layerxs-enterprise.html), read 12 Nov 2026) - LayerX says the browser extension runs on Chrome, Edge, Firefox and Safari on managed and unmanaged devices, and the native desktop agent installs through standard tooling; no operating systems or install time are listed. (the vendor's own claim; [manilatimes.net](https://www.manilatimes.net/2026/05/21/tmt-newswire/globenewswire/layerx-security-integrates-with-new-extended-plan-for-aws-security-hub/2348650), read 13 Nov 2026) ## Checks that found nothing - Any statement that Workforce Protector inventories IDE assistants, command-line coding agents (Claude Code, Codex CLI, Gemini CLI) or local MCP servers on the endpoint: none found on 12 Nov 2026 (checked: [akamai.com](https://www.akamai.com/products/akamai-workforce-protector)) - Documented blocking of shell commands, local file reads, outbound network calls or credential access by AI agents running outside the browser: none found on 12 Nov 2026 (checked: [akamai.com](https://www.akamai.com/solutions/ai-usage-control)) - A published time-to-deploy figure or an operating-system support list for the desktop visibility component on the product page: none found on 12 Nov 2026 (checked: [akamai.com](https://www.akamai.com/products/akamai-workforce-protector)) - A third-party or observed source (analyst note, independent test, public docs) confirming LayerX policy across browsers, SaaS copilots and cloud workspaces beyond vendor statements; the news searches returned vendor releases and acquisition coverage only: none found on 13 Nov 2026 (checked: [news.google.com](https://news.google.com/rss/search?q=%22LayerX%22+browser+OR+SaaS+OR+copilot)) - A published price for Workforce Protector; the Akamai page offers a free demo form only: none found on 12 Nov 2026 (checked: [akamai.com](https://www.akamai.com/products/akamai-workforce-protector)) - Public documentation; docs.layerxsecurity.com redirects to a login page and techdocs.akamai.com/workforce-protector redirects to Akamai Control Center sign-in: none found on 12 Nov 2026 (checked: [docs.layerxsecurity.com](https://docs.layerxsecurity.com)) - Akamai's Workforce Protector docs at techdocs.akamai.com redirect to the Akamai Control Center login, so no public product documentation was readable (consistent with layerx-25): none found on 12 Nov 2026 (checked: [techdocs.akamai.com](https://techdocs.akamai.com/workforce-protector/docs/welcome)) ## Position in each situation | Situation | Position | Total | Leader | | --- | --- | --- | --- | | [Coding agents](/best-for/coding-agents/) | 15th of 16 | 50.5 | Bay 75.5 | | [Regulated data](/best-for/regulated-data/) | 13th of 16 | 52.0 | Microsoft 79.0 | | [Lean team with EDR](/best-for/lean-team-edr/) | 15th of 16 | 50.5 | Bay 76.5 | | [Staff rollout](/best-for/staff-rollout/) | 13th of 16 | 50.5 | Microsoft 75.5 | | [Board in 30 days](/best-for/board-30-days/) | 15th of 16 | 50.0 | Bay 77.5 | LayerX under each situation's own weights. Tied marks an equal total; the order follows the tie-break, the higher score on the heaviest criterion, then name. ## Head-to-heads - [Island vs LayerX](/compare/island-vs-layerx/): Island 66.0 to LayerX 50.0 on the published weights **About Bay.** Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). In regulated data, Bay ranks 3rd of 16 at 70.5. In staff rollout, Bay ranks 2nd of 16 at 73.5. Bay is a young company that left stealth in 2026, publishes no price and has little independent testing behind its claims. More: [all 16 platform profiles](/platforms/), [every head-to-head](/compare/), [the six agent surfaces](/surfaces/) and [the six agent actions](/actions/). ## Sources - www.akamai.com/products/akamai-workforce-protector: https://www.akamai.com/products/akamai-workforce-protector (read 12 Nov 2026) - thehackernews.com/2023/11/hands-on-review-layerxs-enterprise.html: https://thehackernews.com/2023/11/hands-on-review-layerxs-enterprise.html (read 12 Nov 2026) - www.akamai.com/blog/security/layerx-announces-the-first-dedicated-solution-for-agentic-browser-protection: https://www.akamai.com/blog/security/layerx-announces-the-first-dedicated-solution-for-agentic-browser-protection (read 12 Nov 2026) - www.manilatimes.net/2026/05/21/tmt-newswire/globenewswire/layerx-security-integrates-with-new-extended-plan-for-aws-security-hub/2348650: https://www.manilatimes.net/2026/05/21/tmt-newswire/globenewswire/layerx-security-integrates-with-new-extended-plan-for-aws-security-hub/2348650 (read 13 Nov 2026) - www.akamai.com/solutions/ai-usage-control (checked, nothing found): https://www.akamai.com/solutions/ai-usage-control (read 12 Nov 2026) - www.akamai.com/newsroom/press-release/akamai-technologies-announces-intent-to-acquire-layerx-advancing-its-workforce-security-strategy-with-ai-usage-control: https://www.akamai.com/newsroom/press-release/akamai-technologies-announces-intent-to-acquire-layerx-advancing-its-workforce-security-strategy-with-ai-usage-control (read 12 Nov 2026) - news.google.com/rss/search?q=%22LayerX%22+browser+OR+SaaS+OR+copilot (checked, nothing found): https://news.google.com/rss/search?q=%22LayerX%22+browser+OR+SaaS+OR+copilot (read 13 Nov 2026) - www.akamai.com/blog/security-research/claude-desktop-extensions-rce: https://www.akamai.com/blog/security-research/claude-desktop-extensions-rce (read 12 Nov 2026) - www.akamai.com/blog/security/cursorjacking-every-cursor-user-is-vulnerable-to-api-key-theft-by-rogue-extensions: https://www.akamai.com/blog/security/cursorjacking-every-cursor-user-is-vulnerable-to-api-key-theft-by-rogue-extensions (read 12 Nov 2026) - thehackernews.com/2025/10/new-chatgpt-atlas-browser-exploit-lets.html: https://thehackernews.com/2025/10/new-chatgpt-atlas-browser-exploit-lets.html (read 12 Nov 2026) - www.infosecurity-magazine.com/news/bioshocking-ai-browser-prompt: https://www.infosecurity-magazine.com/news/bioshocking-ai-browser-prompt/ (read 13 Nov 2026) - docs.layerxsecurity.com (checked, nothing found): https://docs.layerxsecurity.com (read 12 Nov 2026) - techdocs.akamai.com/workforce-protector/docs/welcome (checked, nothing found): https://techdocs.akamai.com/workforce-protector/docs/welcome (read 12 Nov 2026) - www.securityweek.com/layerx-raises-26-million-for-browser-security-platform: https://www.securityweek.com/layerx-raises-26-million-for-browser-security-platform/ (read 12 Nov 2026) - www.securityweek.com/layerx-raises-75m-seed-funding-tackle-secure-web-browsing: https://www.securityweek.com/layerx-raises-75m-seed-funding-tackle-secure-web-browsing/ (read 12 Nov 2026) - www.akamai.com/blog/security/layerx-joins-akamai-to-scale-up-ai-usage-control: https://www.akamai.com/blog/security/layerx-joins-akamai-to-scale-up-ai-usage-control (read 12 Nov 2026) - www.akamai.com/blog/ai/ai-usage-control-employee-ai-use-needs-own-category: https://www.akamai.com/blog/ai/ai-usage-control-employee-ai-use-needs-own-category (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/platforms/layerx/ --- # Palo Alto Networks: 16th of 16 at 45.0, strongest on Proof (named customers, funding, analyst coverage) at 8 of 10 PLATFORM PROFILE, read 13 Nov 2026 **The short answer.** Palo Alto Networks ranks 16th of 16 AI endpoint security tools at 45.0 of 100. Palo Alto Networks stands out on Proof (named customers, funding, analyst coverage) at 8 of 10, and its lowest score is Sees at 3 of 10. Scores run 0 to 100 on weights fixed before any vendor was scored. ## Profile - **Owner:** Palo Alto Networks, Inc. - **Product:** Prisma AIRS (AI Runtime Security) with Prisma Browser; endpoint agent controls via Cortex Agentic Endpoint Security - **Status:** Active under this name, checked 13 Nov 2026 - **Category:** Platform security vendor: AI runtime security, enterprise browser and agentic endpoint security - **Deployment:** Prisma AIRS: Network Intercept and API Intercept modes; Prisma Browser: desktop browser, extension or mobile app (vendor). Cortex Agentic Endpoint Security is profiled separately as Koi. - **Operating systems:** none found on 12 Nov 2026 - **Time to deploy:** none found on 12 Nov 2026 - **Price:** none found on 12 Nov 2026 - **Website:** [paloaltonetworks.com](https://www.paloaltonetworks.com/prisma/prisma-ai-runtime-security) Scores out of 10 (weight)Sees 253Stops 255Deployment 154Beyond 106Policy 105Transparency 104Proof 58 ## Strength and weakness **Strength:** Palo Alto Networks documents both a browser-level control point (extension inventory, copy-paste and upload limits for unsanctioned AI apps) and endpoint coding-agent policy enforcement under Prisma AIRS, backed by public docs and analyst recognition. **Weakness:** None found on 12 Nov 2026: a public list of which endpoint coding-agent actions are blocked before completion, supported OS, deployment time, or a published price. ## Scores: 45.0 of 100, 16th of 16 | Criterion | Weight | Score | Why this score | | --- | --- | --- | --- | | Sees agentic activity on the endpoint | 25 | 3 | The vendor documents an inventory of browser extensions and claims visibility into all AI apps and agentic workflows; observed docs show an AI agent discovery feature with first-phase limits on cloud data. Named desktop agents, IDE assistants, command-line agents or an endpoint MCP inventory were none found on 13 Nov 2026. | | Stops the action | 25 | 5 | The vendor blocks pastes, uploads and transfers to unsanctioned apps and claims coding-agent enforcement that helps stop destructive commands; trade press confirms network policy enforcement on DPUs is shipping, while a CPU gateway is forward-looking. A public page naming which agent actions are blocked before completion was none found on 12 Nov 2026. | | Deployment friction | 15 | 4 | The vendor ships a browser, extension or mobile app deployable from an email link without admin rights, plus network and API intercept modes. An operating-system list, endpoint agent footprint or time to deploy was none found on 13 Nov 2026. | | Beyond the endpoint | 10 | 6 | The vendor says one browser policy controls SaaS, GenAI and private apps, and claims coverage across cloud, SaaS and endpoints, with SaaS agent prompts monitored rather than blocked. A partner integration for agent traffic has no release date, and an independent confirmation of one shared policy was none found on 13 Nov 2026. | | Policy and governance | 10 | 5 | The vendor describes dynamic risk-based policies and audit trails across web actions, and trade press reports an integration that enforces network and API access rules and sends telemetry to the vendor's own SIEM platform. A named EDR or identity-provider integration was none found on 13 Nov 2026. | | Transparency | 10 | 4 | Public technical documentation, including a dated browser product docs site, and release notes were observed. A published price and published research were none found on 12 Nov 2026. | | Proof | 5 | 8 | Third-party press confirms an acquisition of more than $500 million, the vendor names customer voices at three companies, and it cites a Gartner placement. The analyst placement is vendor-stated. | Palo Alto Networks: weighted total 45.0 of 100 on the published weights. ## What does Palo Alto Networks see on an endpoint? Palo Alto Networks documents 2 of 6 agent surfaces in full (IDE assistants and browser extensions) and 3 in part (desktop apps, command-line agents and SaaS copilots). Nothing was found for MCP servers. Each cell with its note and source: [Palo Alto Networks on the Agent Surface Map](/agent-surface-map/palo-alto-networks/). ## Which agent actions can Palo Alto Networks stop? Palo Alto Networks documents stopping 1 of 6 agent actions before they complete (data paste) and 2 in part (shell command and network call). Nothing was found for file read, credential access and package install. Each cell with its note and source: [Palo Alto Networks on the Action Control Matrix](/action-control-matrix/palo-alto-networks/). ## How much does Palo Alto Networks cost? Palo Alto Networks publishes no list price. None found on 12 Nov 2026. Price counts toward no score; deployment friction is scored under Deployment. ## Evidence behind the scores - Palo Alto Networks says Prisma AIRS discovers AI apps and agents everywhere and governs agents at scale. (the vendor's own claim; [paloaltonetworks.com](https://www.paloaltonetworks.com/prisma/prisma-ai-runtime-security), read 12 Nov 2026) - Prisma Browser inventories browser extensions in use and blocks risky or over-permissioned ones. (the vendor's own claim; [paloaltonetworks.com](https://www.paloaltonetworks.com/sase/prisma-browser), read 12 Nov 2026) - Palo Alto Networks claims Prisma Browser gives visibility into all GenAI apps and agentic workflows. (the vendor's own claim; [paloaltonetworks.com](https://www.paloaltonetworks.com/sase/prisma-browser), read 12 Nov 2026) - Palo Alto Networks documentation limits Prisma AIRS AI Agent Discovery: Azure model logs are unsupported in the initial phase, so Azure shows static configuration without runtime invocation data. (observed in public documentation or code; [docs.paloaltonetworks.com](https://docs.paloaltonetworks.com/ai-runtime-security/administration/agent-discovery/ai-agent-discovery-limitations), read 13 Nov 2026) - Palo Alto Networks documentation limits Prisma AIRS AI Agent Discovery: it gives no threat information before or after protection is applied in this first phase. (observed in public documentation or code; [docs.paloaltonetworks.com](https://docs.paloaltonetworks.com/ai-runtime-security/administration/agent-discovery/ai-agent-discovery-limitations), read 13 Nov 2026) - A Prisma AIRS release note dated September 2026 announces Prisma AIRS Runtime plus AES as generally available, with policy enforcement for AI coding agents on endpoints. (the vendor's own claim; [paloaltonetworks.com](https://www.paloaltonetworks.com/prisma/prisma-ai-runtime-security), read 12 Nov 2026) - Palo Alto Networks says Cortex XDR One enforces runtime guardrails in the agent layer to help stop prompt injections and destructive commands, and blocks exploits and living-off-the-land attacks at the OS layer (October 2026). (the vendor's own claim; [paloaltonetworks.com](https://www.paloaltonetworks.com/blog/2026/10/reinventing-endpoint-security-with-cortex-xdr-one/), read 12 Nov 2026) - Prisma Browser can redact sensitive data in prompts and restrict copy-paste or file uploads to unsanctioned AI apps. (the vendor's own claim; [paloaltonetworks.com](https://www.paloaltonetworks.com/sase/prisma-browser), read 12 Nov 2026) - Prisma Browser says it prevents leakage by blocking transfers between sanctioned corporate apps and personal accounts. (the vendor's own claim; [paloaltonetworks.com](https://www.paloaltonetworks.com/sase/prisma-browser), read 12 Nov 2026) - Help Net Security reports Prisma AIRS AI Runtime Security on NVIDIA BlueField DPUs is generally available and enforces network policy separately from the host running the agent. (third-party source; [helpnetsecurity.com](https://www.helpnetsecurity.com/2026/09/29/palo-alto-networks-nvidia-ai-agent-security/), read 13 Nov 2026) - Limit reported by Help Net Security: the Prisma AIRS AI Gateway on NVIDIA Vera CPUs is described by the company as forward-looking, not shipping. (third-party source; [helpnetsecurity.com](https://www.helpnetsecurity.com/2026/09/29/palo-alto-networks-nvidia-ai-agent-security/), read 13 Nov 2026) - Prisma Browser ships as a dedicated desktop enterprise browser, a browser extension for existing browsers, and a mobile app. (the vendor's own claim; [paloaltonetworks.com](https://www.paloaltonetworks.com/sase/prisma-browser), read 12 Nov 2026) - Palo Alto Networks says Prisma Browser can be deployed from an email link without admin privileges. (the vendor's own claim; [paloaltonetworks.com](https://www.paloaltonetworks.com/sase/prisma-browser), read 12 Nov 2026) - Prisma AIRS lists Network Intercept and API Intercept as its deployment modes. (the vendor's own claim; [paloaltonetworks.com](https://www.paloaltonetworks.com/prisma/prisma-ai-runtime-security), read 12 Nov 2026) ## Checks that found nothing - An independent (third-party or observed) source that documents Cortex AES or Prisma AIRS endpoint inventory of AI agents, MCP servers or extensions; Google News query C1 returned only Palo Alto Networks own pages: none found on 13 Nov 2026 (checked: [news.google.com](https://news.google.com/rss/search?q=%22Palo+Alto+Networks%22+(inventory+OR+discovery+OR+MCP+OR+%22coding+agents%22))) - A public page naming which coding-agent actions (shell command, file read, credential access) Prisma AIRS endpoint enforcement blocks before completion: none found on 12 Nov 2026 (checked: [paloaltonetworks.com](https://www.paloaltonetworks.com/prisma/prisma-ai-runtime-security)) - Operating systems, endpoint agent footprint or time to deploy for the Prisma AIRS and AES endpoint components: none found on 12 Nov 2026 (checked: [paloaltonetworks.com](https://www.paloaltonetworks.com/prisma/prisma-ai-runtime-security)) - An independent source giving operating systems, agent footprint or time to deploy for Cortex AES or Prisma AIRS endpoint enforcement; Google News query C3 and the AES product page returned none: none found on 13 Nov 2026 (checked: [news.google.com](https://news.google.com/rss/search?q=%22Palo+Alto+Networks%22+(agentless+OR+deploy+OR+macOS+OR+Windows))) - An independent source confirming that Prisma AIRS, Prisma Browser and Cortex AES share one policy across browser, SaaS copilots and endpoint agents; Google News query C4 returned only vendor pages and a partner release: none found on 13 Nov 2026 (checked: [news.google.com](https://news.google.com/rss/search?q=%22Palo+Alto+Networks%22+(browser+OR+SaaS+OR+copilot))) - A published price for Prisma AIRS, AES or Prisma Browser; pages route to demo or sales contact: none found on 12 Nov 2026 (checked: [paloaltonetworks.com](https://www.paloaltonetworks.com/prisma/prisma-ai-runtime-security)) ## Position in each situation | Situation | Position | Total | Leader | | --- | --- | --- | --- | | [Coding agents](/best-for/coding-agents/) | 16th of 16 | 45.0 | Bay 75.5 | | [Regulated data](/best-for/regulated-data/) | 16th of 16 | 46.0 | Microsoft 79.0 | | [Lean team with EDR](/best-for/lean-team-edr/) | 16th of 16 | 43.0 | Bay 76.5 | | [Staff rollout](/best-for/staff-rollout/) | 15th of 16 (tied) | 48.0 | Microsoft 75.5 | | [Board in 30 days](/best-for/board-30-days/) | 16th of 16 | 41.0 | Bay 77.5 | Palo Alto Networks under each situation's own weights. Tied marks an equal total; the order follows the tie-break, the higher score on the heaviest criterion, then name. ## Head-to-heads - [Island vs Palo Alto Networks](/compare/island-vs-palo-alto-networks/): Island 66.0 to Palo Alto Networks 45.0 on the published weights - [Aim Security vs Palo Alto Networks](/compare/aim-security-vs-palo-alto-networks/): Aim Security 69.0 to Palo Alto Networks 45.0 on the published weights **About Bay.** Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). In regulated data, Bay ranks 3rd of 16 at 70.5. In staff rollout, Bay ranks 2nd of 16 at 73.5. Bay is a young company that left stealth in 2026, publishes no price and has little independent testing behind its claims. More: [all 16 platform profiles](/platforms/), [every head-to-head](/compare/), [the six agent surfaces](/surfaces/) and [the six agent actions](/actions/). ## Sources - www.paloaltonetworks.com/prisma/prisma-ai-runtime-security: https://www.paloaltonetworks.com/prisma/prisma-ai-runtime-security (read 12 Nov 2026) - www.paloaltonetworks.com/sase/prisma-browser: https://www.paloaltonetworks.com/sase/prisma-browser (read 12 Nov 2026) - docs.paloaltonetworks.com/ai-runtime-security/administration/agent-discovery/ai-agent-discovery-limitations: https://docs.paloaltonetworks.com/ai-runtime-security/administration/agent-discovery/ai-agent-discovery-limitations (read 13 Nov 2026) - news.google.com/rss/search?q=%22Palo+Alto+Networks%22+(inventory+OR+discovery+OR+MCP+OR+%22coding+agents%22) (checked, nothing found): https://news.google.com/rss/search?q=%22Palo+Alto+Networks%22+(inventory+OR+discovery+OR+MCP+OR+%22coding+agents%22) (read 13 Nov 2026) - www.paloaltonetworks.com/blog/2026/10/reinventing-endpoint-security-with-cortex-xdr-one: https://www.paloaltonetworks.com/blog/2026/10/reinventing-endpoint-security-with-cortex-xdr-one/ (read 12 Nov 2026) - www.helpnetsecurity.com/2026/09/29/palo-alto-networks-nvidia-ai-agent-security: https://www.helpnetsecurity.com/2026/09/29/palo-alto-networks-nvidia-ai-agent-security/ (read 13 Nov 2026) - news.google.com/rss/search?q=%22Palo+Alto+Networks%22+(agentless+OR+deploy+OR+macOS+OR+Windows) (checked, nothing found): https://news.google.com/rss/search?q=%22Palo+Alto+Networks%22+(agentless+OR+deploy+OR+macOS+OR+Windows) (read 13 Nov 2026) - www.paloaltonetworks.com/blog/2026/10/palo-alto-networks-brings-continuous-ai-defense-and-prisma-airs-to-the-openai-marketplace: https://www.paloaltonetworks.com/blog/2026/10/palo-alto-networks-brings-continuous-ai-defense-and-prisma-airs-to-the-openai-marketplace/ (read 13 Nov 2026) - www.paloaltonetworks.com/blog/2026/11/new-cortex-xdr-integration-with-claude-cowork-secures-desktop-ai-agents: https://www.paloaltonetworks.com/blog/2026/11/new-cortex-xdr-integration-with-claude-cowork-secures-desktop-ai-agents/ (read 13 Nov 2026) - siliconangle.com/2026/08/20/zero-networks-expands-palo-alto-networks-integration-to-ai-agent-control: https://siliconangle.com/2026/08/20/zero-networks-expands-palo-alto-networks-integration-to-ai-agent-control/ (read 13 Nov 2026) - news.google.com/rss/search?q=%22Palo+Alto+Networks%22+(browser+OR+SaaS+OR+copilot) (checked, nothing found): https://news.google.com/rss/search?q=%22Palo+Alto+Networks%22+(browser+OR+SaaS+OR+copilot) (read 13 Nov 2026) - docs.paloaltonetworks.com/ai-runtime-security: https://docs.paloaltonetworks.com/ai-runtime-security (read 12 Nov 2026) - docs.paloaltonetworks.com/ai-runtime-security/release-notes: https://docs.paloaltonetworks.com/ai-runtime-security/release-notes (read 12 Nov 2026) - docs.paloaltonetworks.com/prisma-access-browser: https://docs.paloaltonetworks.com/prisma-access-browser (read 13 Nov 2026) - siliconangle.com/2025/08/05/sentinelone-acquires-ai-security-startup-prompt-security: https://siliconangle.com/2025/08/05/sentinelone-acquires-ai-security-startup-prompt-security/ (read 12 Nov 2026) - www.paloaltonetworks.com/company/press/2026/palo-alto-networks-completes-acquisition-of-koi-to-secure-the-agentic-endpoint: https://www.paloaltonetworks.com/company/press/2026/palo-alto-networks-completes-acquisition-of-koi-to-secure-the-agentic-endpoint (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/platforms/palo-alto-networks/ --- # 16 AI endpoint security platforms, each profiled from its own documentation PLATFORMS, read 13 Nov 2026 **The short answer.** Profiles here cover 16 AI endpoint security platforms, from Microsoft at 75.0 of 100 to Palo Alto Networks at 45.0. Each profile lists the owner, deployment model, price as published, seven scores with reasons and the facts behind them, with sources and read dates. - [1. Microsoft](/platforms/microsoft/): 75.0 of 100. Defender discovers dozens of local agents and MCP servers on Windows and macOS (GA) and can block prompt-injection-driven actions in Claude Code, Codex CLI and GitHub Copilot CLI on Windows, with Endpoint DLP blocking pastes and uploads to AI sites. - [2. Bay](/platforms/bay/): 71.5 of 100. Bay documents on-device, session-aware allow/ask/deny/steer enforcement with named action types, agentless MDM/EDR deployment in under 5 minutes (Bay's claim), and one policy across endpoint, browser and cloud. - [3. Harmonic Security](/platforms/harmonic-security/): 71.5 of 100. Harmonic documents hook-based inline blocking of prompts and tool calls across named coding agents (Claude Code, Codex, Cursor, Copilot CLI and others) plus an MCP gateway with identity-aware access groups. - [4. Aim Security](/platforms/aim-security/): 69.0 of 100. Cato AI Endpoint documents a lightweight client that finds local agents with their MCP servers and skills and blocks an unapproved tool call before it runs, in one dashboard with inline network inspection. - [5. CrowdStrike](/platforms/crowdstrike/): 66.0 of 100. Falcon Guardian pairs OS-level sensor telemetry with agent-level hooks that can block actions before they run in Claude Code, Codex, GitHub Copilot, OpenCode and Cursor, plus package blocking in the same sensor. - [6. Noma Security](/platforms/noma-security/): 66.0 of 100. Has a dedicated, current endpoint page: discovery of agents, MCP servers and skills via existing EDR or MDM, plus action-level allow and block policy tied to IdP identity. - [7. Island](/platforms/island/): 66.0 of 100. Island combines an endpoint agent for Windows and macOS with a browser, extension and MCP gateway, and documents blocking of Bash execution, file reads and network connections as well as clipboard DLP. - [8. Zenity](/platforms/zenity/): 65.0 of 100. Documents pre-execution blocking of coding-agent tool calls through native hooks and an MCP gateway, including credential-read-then-network sequences. - [9. Knostic](/platforms/knostic/): 64.0 of 100. Knostic documents Enforce-mode blocking of secret file reads, risky MCP servers, destructive commands and unsafe packages inside agents, with a published $15 per user Team price and free tier. - [10. SentinelOne](/platforms/sentinelone/): 62.0 of 100. SentinelOne documents an MCP gateway with allow, block, filter or redact actions and runtime tool-call blocking, plus a lightweight agent and extension that discover shadow AI across browsers, IDEs and terminal assistants. - [11. WitnessAI](/platforms/witnessai/): 61.0 of 100. WitnessAI documents inline blocking of unapproved MCP servers and tool calls across IDEs, chat apps and custom agents, with audit records per blocked call and one policy for humans and agents. - [12. Koi](/platforms/koi/): 54.5 of 100. Palo Alto Networks documents inventory of IDE plugins, browser extensions and MCP servers on endpoints, plus real-time blocking of risky downloads and agent actions, with standalone or Cortex XDR deployment. - [13. Lasso Security](/platforms/lasso-security/): 53.0 of 100. Lasso documents inspecting every coding-agent tool call before execution via Claude Code, Cursor and Codex hooks, with out-of-scope calls flagged or blocked, plus an open-source MCP gateway. - [14. Pillar Security](/platforms/pillar-security/): 53.0 of 100. Documents pre-action blocking in the agent's own pre-tool hook for Claude Code, Codex CLI and Cursor, with credential-exfiltration and persistence controls and MDM-installed hooks developers cannot disable. - [15. LayerX](/platforms/layerx/): 50.0 of 100. It enforces warn, redact, block or guide policy on prompts, uploads and copy-paste at the moment of interaction with no proxy or browser replacement, and covers agentic browsers. - [16. Palo Alto Networks](/platforms/palo-alto-networks/): 45.0 of 100. Palo Alto Networks documents both a browser-level control point (extension inventory, copy-paste and upload limits for unsanctioned AI apps) and endpoint coding-agent policy enforcement under Prisma AIRS, backed by public docs and analyst recognition. **About Bay.** Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). ## Sources - Scores dataset, CSV: https://aiendpointsecurityranked.com/datasets/aes-scores.csv (read 13 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/platforms/ --- # Bay beats Island 71.5 to 66.0 on the published weights HEAD TO HEAD, read 13 Nov 2026 **The short answer.** Bay scores 71.5 of 100 against Island at 66.0 on the published weights. Bay leads on Deployment (9 to 6), Sees (8 to 7) and Stops (8 to 7), while Island is ahead on Transparency (3 to 1), Beyond (8 to 7) and Policy (7 to 6). By situation, Bay leads in all five. ## Verdict on the published weights Bay scores 71.5 of 100 and Island scores 66.0. Bay is ahead on 3 of 7 criteria and behind on 3; 1 is level. Bay leads on Deployment (9 to 6), Sees (8 to 7) and Stops (8 to 7), while Island is ahead on Transparency (3 to 1), Beyond (8 to 7) and Policy (7 to 6). **Where Bay is behind.** Bay is a young company that left stealth in 2026, publishes no price and has little independent testing behind its claims. ## Criterion by criterion | Criterion | Weight | Bay | Island | | --- | --- | --- | --- | | Sees agentic activity on the endpoint | 25 | 8 | 7 | | Stops the action | 25 | 8 | 7 | | Deployment friction | 15 | 9 | 6 | | Beyond the endpoint | 10 | 7 | 8 | | Policy and governance | 10 | 6 | 7 | | Transparency | 10 | 1 | 3 | | Proof | 5 | 8 | 8 | | Total (0 to 100) | 100 | 71.5 | 66.0 | Bay against Island, scores out of 10 on the site weights. ## What each sees and can stop | Tool | Surface strip | Stop strip | | --- | --- | --- | | [Bay](/platforms/bay/) | Surfaces seen: Desktop apps partial; IDE assistants documented; Command-line agents documented; MCP servers documented; Browser extensions documented; SaaS copilots partial | Actions it can stop: File read partial; Shell command documented; Network call documented; Credential access documented; Data paste partial; Package install partial | | [Island](/platforms/island/) | Surfaces seen: Desktop apps documented; IDE assistants documented; Command-line agents partial; MCP servers documented; Browser extensions documented; SaaS copilots partial | Actions it can stop: File read documented; Shell command documented; Network call documented; Credential access partial; Data paste documented; Package install partial | Filled documented, half partial, open none found. Surface order: desktop apps, IDE assistants, command-line agents, MCP servers, browser extensions, SaaS copilots. Stop order: file read, shell command, network call, credential access, data paste, package install. | Where they differ | Bay | Island | | --- | --- | --- | | Desktop apps | partial | documented | | Command-line agents | documented | partial | | Stops: file read | partial | documented | | Stops: credential access | documented | partial | | Stops: data paste | partial | documented | 5 of 12 cells differ between Bay and Island as documented on 13 Nov 2026. ## Who should pick what, by situation - **Coding agents** ([an engineering organization where every developer runs coding agents with shell access](/best-for/coding-agents/)): Bay leads Island 75.5 to 67.5, ahead on Deployment (9 to 6), Stops (8 to 7) and Sees (8 to 7), while Island is ahead on Policy (7 to 6), Transparency (3 to 1) and Beyond (8 to 7). - **Regulated data** ([a regulated company that must show which AI tools touch customer data](/best-for/regulated-data/)): Bay leads Island 70.5 to 69.5, ahead on Sees (8 to 7), Stops (8 to 7) and Deployment (9 to 6), while Island is ahead on Policy (7 to 6), Beyond (8 to 7) and Transparency (3 to 1). - **Lean team with EDR** ([a lean security team with EDR in place and no view of AI agents](/best-for/lean-team-edr/)): Bay leads Island 76.5 to 66.5, ahead on Deployment (9 to 6), Sees (8 to 7) and Stops (8 to 7), while Island is ahead on Policy (7 to 6), Transparency (3 to 1) and Beyond (8 to 7). - **Staff rollout** ([a company rolling out an AI assistant and browser agents to all staff](/best-for/staff-rollout/)): Bay leads Island 73.5 to 69.5, ahead on Deployment (9 to 6), Sees (8 to 7) and Stops (8 to 7), while Island is ahead on Beyond (8 to 7), Policy (7 to 6) and Transparency (3 to 1). - **Board in 30 days** ([a CISO asked by the board for an inventory of AI agents within 30 days](/best-for/board-30-days/)): Bay leads Island 77.5 to 66.5, ahead on Deployment (9 to 6), Sees (8 to 7) and Stops (8 to 7), while Island is ahead on Beyond (8 to 7), Transparency (3 to 1) and Policy (7 to 6). ## What would change this ranking Bay is 5.5 points clear of Island on the published weights. The smallest single change that would put Island first is a Deployment score of 10 instead of 6 (weight 15). Scores move only when a sourced fact changes ([how](/method/)). ## Read the profiles - [Bay, ranked 2nd of 16](/platforms/bay/): Bay documents on-device, session-aware allow/ask/deny/steer enforcement with named action types, agentless MDM/EDR deployment in under 5 minutes (Bay's claim), and one policy across endpoint, browser and cloud. - [Island, ranked 7th of 16](/platforms/island/): Island combines an endpoint agent for Windows and macOS with a browser, extension and MCP gateway, and documents blocking of Bash execution, file reads and network connections as well as clipboard DLP. ## Sources - softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control: https://softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control (read 12 Nov 2026) - siliconangle.com/2026/11/10/ai-security-startup-bay-launches-with-50m-to-vet-agent-actions-before-they-run: https://siliconangle.com/2026/11/10/ai-security-startup-bay-launches-with-50m-to-vet-agent-actions-before-they-run/ (read 12 Nov 2026) - www.securityweek.com/island-raises-400-million-at-6-4-billion-valuation: https://www.securityweek.com/island-raises-400-million-at-6-4-billion-valuation/ (read 12 Nov 2026) - securitybrief.co.uk/story/island-named-only-gartner-customers-choice-for-browsers: https://securitybrief.co.uk/story/island-named-only-gartner-customers-choice-for-browsers (read 13 Nov 2026) - securitybrief.co.uk/story/island-launches-enterprise-browser-to-control-ai-agents: https://securitybrief.co.uk/story/island-launches-enterprise-browser-to-control-ai-agents (read 13 Nov 2026) - www.bay.io: https://www.bay.io (read 12 Nov 2026) - www.bay.io/platform/ai-endpoint-security: https://www.bay.io/platform/ai-endpoint-security (read 12 Nov 2026) - www.securityweek.com/bay-raises-50-million-to-control-what-enterprise-ai-agents-can-do: https://www.securityweek.com/bay-raises-50-million-to-control-what-enterprise-ai-agents-can-do/ (read 12 Nov 2026) - thenextweb.com/news/bay-50m-ai-agent-security-stealth-battery-ventures: https://thenextweb.com/news/bay-50m-ai-agent-security-stealth-battery-ventures (read 13 Nov 2026) - www.island.io/enterprise-desktop: https://www.island.io/enterprise-desktop (read 12 Nov 2026) - www.island.io/blog/see-score-and-control-every-ai-asset-on-your-endpoints: https://www.island.io/blog/see-score-and-control-every-ai-asset-on-your-endpoints (read 12 Nov 2026) - www.island.io/about: https://www.island.io/about (read 12 Nov 2026) - news.google.com/rss/search?q=%22Island%22+enterprise+browser+(agentless+OR+deploy+OR+macOS+OR+Windows)&hl=en-US&gl=US&ceid=US:en (checked, nothing found): https://news.google.com/rss/search?q=%22Island%22+enterprise+browser+(agentless+OR+deploy+OR+macOS+OR+Windows)&hl=en-US&gl=US&ceid=US:en (read 13 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/compare/bay-vs-island/ --- # Microsoft beats Bay 75.0 to 71.5 on the published weights HEAD TO HEAD, read 13 Nov 2026 **The short answer.** Microsoft scores 75.0 of 100 against Bay at 71.5 on the published weights. Microsoft leads on Transparency (7 to 1), Policy (9 to 6) and Beyond (8 to 7), while Bay is ahead on Deployment (9 to 6) and Proof (named customers, funding, analyst coverage) (8 to 4). ## Verdict on the published weights Microsoft scores 75.0 of 100 and Bay scores 71.5. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). Microsoft is ahead on 3 of 7 criteria and behind on 2; 2 are level. Microsoft leads on Transparency (7 to 1), Policy (9 to 6) and Beyond (8 to 7), while Bay is ahead on Deployment (9 to 6) and Proof (named customers, funding, analyst coverage) (8 to 4). **Where Bay is behind.** Bay is a young company that left stealth in 2026, publishes no price and has little independent testing behind its claims. ## Criterion by criterion | Criterion | Weight | Bay | Microsoft | | --- | --- | --- | --- | | Sees agentic activity on the endpoint | 25 | 8 | 8 | | Stops the action | 25 | 8 | 8 | | Deployment friction | 15 | 9 | 6 | | Beyond the endpoint | 10 | 7 | 8 | | Policy and governance | 10 | 6 | 9 | | Transparency | 10 | 1 | 7 | | Proof | 5 | 8 | 4 | | Total (0 to 100) | 100 | 71.5 | 75.0 | Bay against Microsoft, scores out of 10 on the site weights. ## What each sees and can stop | Tool | Surface strip | Stop strip | | --- | --- | --- | | [Bay](/platforms/bay/) | Surfaces seen: Desktop apps partial; IDE assistants documented; Command-line agents documented; MCP servers documented; Browser extensions documented; SaaS copilots partial | Actions it can stop: File read partial; Shell command documented; Network call documented; Credential access documented; Data paste partial; Package install partial | | [Microsoft](/platforms/microsoft/) | Surfaces seen: Desktop apps documented; IDE assistants documented; Command-line agents documented; MCP servers documented; Browser extensions partial; SaaS copilots documented | Actions it can stop: File read partial; Shell command documented; Network call partial; Credential access partial; Data paste documented; Package install none found | Filled documented, half partial, open none found. Surface order: desktop apps, IDE assistants, command-line agents, MCP servers, browser extensions, SaaS copilots. Stop order: file read, shell command, network call, credential access, data paste, package install. | Where they differ | Bay | Microsoft | | --- | --- | --- | | Desktop apps | partial | documented | | Browser extensions | documented | partial | | SaaS copilots | partial | documented | | Stops: network call | documented | partial | | Stops: credential access | documented | partial | | Stops: data paste | partial | documented | | Stops: package install | partial | none found | 7 of 12 cells differ between Bay and Microsoft as documented on 13 Nov 2026. ## Who should pick what, by situation - **Coding agents** ([an engineering organization where every developer runs coding agents with shell access](/best-for/coding-agents/)): Bay and Microsoft tie at 75.5 and both score 8 on Stops, the heaviest criterion here, so name order puts Bay 1st. Bay is ahead on Deployment (9 to 6) and Proof (named customers, funding, analyst coverage) (8 to 4), Microsoft on Policy (9 to 6), Transparency (7 to 1) and Beyond (8 to 7). - **Regulated data** ([a regulated company that must show which AI tools touch customer data](/best-for/regulated-data/)): Microsoft leads Bay 79.0 to 70.5 (Bay ranks 3rd of 16 at 70.5), and Microsoft's lead over Bay comes from policy and governance (9 to 6) and transparency (7 to 1). - **Lean team with EDR** ([a lean security team with EDR in place and no view of AI agents](/best-for/lean-team-edr/)): Bay leads Microsoft 76.5 to 73.5, ahead on Deployment (9 to 6) and Proof (named customers, funding, analyst coverage) (8 to 4), while Microsoft is ahead on Policy (9 to 6), Transparency (7 to 1) and Beyond (8 to 7). - **Staff rollout** ([a company rolling out an AI assistant and browser agents to all staff](/best-for/staff-rollout/)): Microsoft leads Bay 75.5 to 73.5 (Bay ranks 2nd of 16 at 73.5), and Microsoft's lead over Bay comes from policy and governance (9 to 6) and transparency (7 to 1). - **Board in 30 days** ([a CISO asked by the board for an inventory of AI agents within 30 days](/best-for/board-30-days/)): Bay leads Microsoft 77.5 to 72.0, ahead on Deployment (9 to 6) and Proof (named customers, funding, analyst coverage) (8 to 4), while Microsoft is ahead on Transparency (7 to 1), Policy (9 to 6) and Beyond (8 to 7). ## What would change this ranking Microsoft is 3.5 points clear of Bay on the published weights. The smallest single change that would put Bay first is a Policy score of 10 instead of 6 (weight 10). Scores move only when a sourced fact changes ([how](/method/)). ## Read the profiles - [Bay, ranked 2nd of 16](/platforms/bay/): Bay documents on-device, session-aware allow/ask/deny/steer enforcement with named action types, agentless MDM/EDR deployment in under 5 minutes (Bay's claim), and one policy across endpoint, browser and cloud. - [Microsoft, ranked 1st of 16](/platforms/microsoft/): Defender discovers dozens of local agents and MCP servers on Windows and macOS (GA) and can block prompt-injection-driven actions in Claude Code, Codex CLI and GitHub Copilot CLI on Windows, with Endpoint DLP blocking pastes and uploads to AI sites. ## Sources - softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control: https://softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control (read 12 Nov 2026) - siliconangle.com/2026/11/10/ai-security-startup-bay-launches-with-50m-to-vet-agent-actions-before-they-run: https://siliconangle.com/2026/11/10/ai-security-startup-bay-launches-with-50m-to-vet-agent-actions-before-they-run/ (read 12 Nov 2026) - learn.microsoft.com/en-us/defender-endpoint/local-agent-discovery-overview: https://learn.microsoft.com/en-us/defender-endpoint/local-agent-discovery-overview (read 12 Nov 2026) - learn.microsoft.com/en-us/defender-endpoint/ai-agent-support-matrix: https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-support-matrix (read 12 Nov 2026) - learn.microsoft.com/en-us/defender-endpoint/whats-new-in-microsoft-defender-endpoint: https://learn.microsoft.com/en-us/defender-endpoint/whats-new-in-microsoft-defender-endpoint (read 12 Nov 2026) - venturebeat.com/technology/microsoft-takes-agent-365-out-of-preview-as-shadow-ai-becomes-an-enterprise-threat: https://venturebeat.com/technology/microsoft-takes-agent-365-out-of-preview-as-shadow-ai-becomes-an-enterprise-threat (read 12 Nov 2026) - www.bay.io: https://www.bay.io (read 12 Nov 2026) - www.bay.io/platform/ai-endpoint-security: https://www.bay.io/platform/ai-endpoint-security (read 12 Nov 2026) - www.securityweek.com/bay-raises-50-million-to-control-what-enterprise-ai-agents-can-do: https://www.securityweek.com/bay-raises-50-million-to-control-what-enterprise-ai-agents-can-do/ (read 12 Nov 2026) - thenextweb.com/news/bay-50m-ai-agent-security-stealth-battery-ventures: https://thenextweb.com/news/bay-50m-ai-agent-security-stealth-battery-ventures (read 13 Nov 2026) - www.bay.io/platform: https://www.bay.io/platform (read 12 Nov 2026) - www.bay.io/ai-agent-investigation: https://www.bay.io/ai-agent-investigation (read 12 Nov 2026) - www.bay.io/ai-governance-compliance: https://www.bay.io/ai-governance-compliance (read 12 Nov 2026) - pulse2.com/bay-50-million-funding-to-expand-ai-agent-security-platform-across-enterprise-environments: https://pulse2.com/bay-50-million-funding-to-expand-ai-agent-security-platform-across-enterprise-environments/ (read 13 Nov 2026) - docs.bay.io (checked, nothing found): https://docs.bay.io (read 13 Nov 2026) - www.bay.io/blog/introducing-bay-the-ai-agent-security-platform-out-of-stealth-with-50m: https://www.bay.io/blog/introducing-bay-the-ai-agent-security-platform-out-of-stealth-with-50m (read 12 Nov 2026) - www.bay.io/sitemap.xml (checked, nothing found): https://www.bay.io/sitemap.xml (read 12 Nov 2026) - www.bay.io/blog (checked, nothing found): https://www.bay.io/blog (read 12 Nov 2026) - www.calcalistech.com/ctechnews/article/bjx11v5x0gx: https://www.calcalistech.com/ctechnews/article/bjx11v5x0gx (read 12 Nov 2026) - learn.microsoft.com/en-us/defender-endpoint/configure-ai-agent-runtime-protection: https://learn.microsoft.com/en-us/defender-endpoint/configure-ai-agent-runtime-protection (read 12 Nov 2026) - learn.microsoft.com/en-us/purview/endpoint-dlp-learn-about: https://learn.microsoft.com/en-us/purview/endpoint-dlp-learn-about (read 12 Nov 2026) - learn.microsoft.com/en-us/defender-endpoint/ai-agent-runtime-protection-overview: https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-runtime-protection-overview (read 12 Nov 2026) - learn.microsoft.com/en-us/purview/dlp-learn-about-dlp: https://learn.microsoft.com/en-us/purview/dlp-learn-about-dlp (read 12 Nov 2026) - learn.microsoft.com/en-us/defender-endpoint/microsoft-defender-endpoint: https://learn.microsoft.com/en-us/defender-endpoint/microsoft-defender-endpoint (read 12 Nov 2026) - learn.microsoft.com/en-us/microsoft-agent-365/overview: https://learn.microsoft.com/en-us/microsoft-agent-365/overview (read 12 Nov 2026) - www.microsoft.com/en-us/microsoft-365/enterprise/microsoft365-plans-and-pricing: https://www.microsoft.com/en-us/microsoft-365/enterprise/microsoft365-plans-and-pricing (read 12 Nov 2026) - www.microsoft.com/en-us/microsoft-agent-365: https://www.microsoft.com/en-us/microsoft-agent-365 (read 12 Nov 2026) - news.google.com/rss/search?q=%22Agent+365%22+price+OR+pricing+per+user&hl=en-US&gl=US&ceid=US:en (checked, nothing found): https://news.google.com/rss/search?q=%22Agent+365%22+price+OR+pricing+per+user&hl=en-US&gl=US&ceid=US:en (read 13 Nov 2026) - www.microsoft.com/en-us/security/blog/2026/05/01/microsoft-agent-365-now-generally-available-expands-capabilities-and-integrations: https://www.microsoft.com/en-us/security/blog/2026/05/01/microsoft-agent-365-now-generally-available-expands-capabilities-and-integrations/ (read 12 Nov 2026) - petri.com/microsoft-agent-365-discover-govern-ai-agents: https://petri.com/microsoft-agent-365-discover-govern-ai-agents/ (read 12 Nov 2026) - www.microsoft.com/en-us/security/business/analyst-reports (checked, nothing found): https://www.microsoft.com/en-us/security/business/analyst-reports (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/compare/bay-vs-microsoft/ --- # Bay beats Knostic 71.5 to 64.0 on the published weights HEAD TO HEAD, read 13 Nov 2026 **The short answer.** Bay scores 71.5 of 100 against Knostic at 64.0 on the published weights. Bay leads most on Beyond (7 to 3), Deployment (9 to 7) and Sees (8 to 7), while Knostic is ahead on Transparency (7 to 1). By situation, Bay leads in all five. ## Verdict on the published weights Bay scores 71.5 of 100 and Knostic scores 64.0. Bay is ahead on 5 of 7 criteria and behind on 1; 1 is level. Bay leads on Beyond (7 to 3), Deployment (9 to 7), Sees (8 to 7), Stops (8 to 7) and Proof (named customers, funding, analyst coverage) (8 to 5), while Knostic is ahead on Transparency (7 to 1). **Where Bay is behind.** Bay is a young company that left stealth in 2026, publishes no price and has little independent testing behind its claims. ## Criterion by criterion | Criterion | Weight | Bay | Knostic | | --- | --- | --- | --- | | Sees agentic activity on the endpoint | 25 | 8 | 7 | | Stops the action | 25 | 8 | 7 | | Deployment friction | 15 | 9 | 7 | | Beyond the endpoint | 10 | 7 | 3 | | Policy and governance | 10 | 6 | 6 | | Transparency | 10 | 1 | 7 | | Proof | 5 | 8 | 5 | | Total (0 to 100) | 100 | 71.5 | 64.0 | Bay against Knostic, scores out of 10 on the site weights. ## What each sees and can stop | Tool | Surface strip | Stop strip | | --- | --- | --- | | [Bay](/platforms/bay/) | Surfaces seen: Desktop apps partial; IDE assistants documented; Command-line agents documented; MCP servers documented; Browser extensions documented; SaaS copilots partial | Actions it can stop: File read partial; Shell command documented; Network call documented; Credential access documented; Data paste partial; Package install partial | | [Knostic](/platforms/knostic/) | Surfaces seen: Desktop apps documented; IDE assistants documented; Command-line agents documented; MCP servers documented; Browser extensions partial; SaaS copilots partial | Actions it can stop: File read documented; Shell command documented; Network call partial; Credential access documented; Data paste partial; Package install partial | Filled documented, half partial, open none found. Surface order: desktop apps, IDE assistants, command-line agents, MCP servers, browser extensions, SaaS copilots. Stop order: file read, shell command, network call, credential access, data paste, package install. | Where they differ | Bay | Knostic | | --- | --- | --- | | Desktop apps | partial | documented | | Browser extensions | documented | partial | | Stops: file read | partial | documented | | Stops: network call | documented | partial | 4 of 12 cells differ between Bay and Knostic as documented on 13 Nov 2026. ## Who should pick what, by situation - **Coding agents** ([an engineering organization where every developer runs coding agents with shell access](/best-for/coding-agents/)): Bay leads Knostic 75.5 to 66.0, ahead on Stops (8 to 7), Deployment (9 to 7), Sees (8 to 7), Beyond (7 to 3) and Proof (named customers, funding, analyst coverage) (8 to 5), while Knostic is ahead on Transparency (7 to 1). - **Regulated data** ([a regulated company that must show which AI tools touch customer data](/best-for/regulated-data/)): Bay leads Knostic 70.5 to 60.5, ahead on Beyond (7 to 3), Sees (8 to 7), Stops (8 to 7), Proof (named customers, funding, analyst coverage) (8 to 5) and Deployment (9 to 7), while Knostic is ahead on Transparency (7 to 1). - **Lean team with EDR** ([a lean security team with EDR in place and no view of AI agents](/best-for/lean-team-edr/)): Bay leads Knostic 76.5 to 66.0, ahead on Deployment (9 to 7), Sees (8 to 7), Stops (8 to 7), Beyond (7 to 3) and Proof (named customers, funding, analyst coverage) (8 to 5), while Knostic is ahead on Transparency (7 to 1). - **Staff rollout** ([a company rolling out an AI assistant and browser agents to all staff](/best-for/staff-rollout/)): Bay leads Knostic 73.5 to 58.0, ahead on Beyond (7 to 3), Deployment (9 to 7), Sees (8 to 7), Stops (8 to 7) and Proof (named customers, funding, analyst coverage) (8 to 5), while Knostic is ahead on Transparency (7 to 1). - **Board in 30 days** ([a CISO asked by the board for an inventory of AI agents within 30 days](/best-for/board-30-days/)): Bay leads Knostic 77.5 to 64.5, ahead on Deployment (9 to 7), Sees (8 to 7), Beyond (7 to 3), Proof (named customers, funding, analyst coverage) (8 to 5) and Stops (8 to 7), while Knostic is ahead on Transparency (7 to 1). ## What would change this ranking Bay is 7.5 points clear of Knostic on the published weights. No single score change could close that gap. Scores move only when a sourced fact changes ([how](/method/)). ## Read the profiles - [Bay, ranked 2nd of 16](/platforms/bay/): Bay documents on-device, session-aware allow/ask/deny/steer enforcement with named action types, agentless MDM/EDR deployment in under 5 minutes (Bay's claim), and one policy across endpoint, browser and cloud. - [Knostic, ranked 9th of 16](/platforms/knostic/): Knostic documents Enforce-mode blocking of secret file reads, risky MCP servers, destructive commands and unsafe packages inside agents, with a published $15 per user Team price and free tier. ## Sources - softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control: https://softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control (read 12 Nov 2026) - siliconangle.com/2026/11/10/ai-security-startup-bay-launches-with-50m-to-vet-agent-actions-before-they-run: https://siliconangle.com/2026/11/10/ai-security-startup-bay-launches-with-50m-to-vet-agent-actions-before-they-run/ (read 12 Nov 2026) - github.com/knostic/openclaw-detect: https://github.com/knostic/openclaw-detect (read 12 Nov 2026) - github.com/knostic/openclaw-shield: https://github.com/knostic/openclaw-shield (read 12 Nov 2026) - siliconangle.com/2025/03/05/knostic-raises-11m-strengthen-enterprise-ai-security-need-know-access-controls: https://siliconangle.com/2025/03/05/knostic-raises-11m-strengthen-enterprise-ai-security-need-know-access-controls/ (read 12 Nov 2026) - github.com/knostic/openclaw-telemetry: https://github.com/knostic/openclaw-telemetry (read 12 Nov 2026) - github.com/knostic/AgentSonar: https://github.com/knostic/AgentSonar (read 12 Nov 2026) - github.com/knostic/OpenAnt: https://github.com/knostic/OpenAnt (read 12 Nov 2026) - www.bay.io/platform/ai-endpoint-security: https://www.bay.io/platform/ai-endpoint-security (read 12 Nov 2026) - www.bay.io/platform/cloud-workspaces: https://www.bay.io/platform/cloud-workspaces (read 12 Nov 2026) - www.bay.io/platform/browser: https://www.bay.io/platform/browser (read 12 Nov 2026) - www.bay.io/blog/bay-emerges-from-stealth-with-50m-after-fortune-100-companies-used-its-platform-to-triple-autonomous-ai-agent-use-2: https://www.bay.io/blog/bay-emerges-from-stealth-with-50m-after-fortune-100-companies-used-its-platform-to-triple-autonomous-ai-agent-use-2 (read 12 Nov 2026) - pulse2.com/bay-50-million-funding-to-expand-ai-agent-security-platform-across-enterprise-environments: https://pulse2.com/bay-50-million-funding-to-expand-ai-agent-security-platform-across-enterprise-environments/ (read 13 Nov 2026) - www.bay.io/blog/introducing-bay-the-ai-agent-security-platform-out-of-stealth-with-50m: https://www.bay.io/blog/introducing-bay-the-ai-agent-security-platform-out-of-stealth-with-50m (read 12 Nov 2026) - docs.bay.io: https://docs.bay.io (read 12 Nov 2026) - www.bay.io/sitemap.xml (checked, nothing found): https://www.bay.io/sitemap.xml (read 12 Nov 2026) - www.bay.io/blog (checked, nothing found): https://www.bay.io/blog (read 12 Nov 2026) - www.securityweek.com/bay-raises-50-million-to-control-what-enterprise-ai-agents-can-do: https://www.securityweek.com/bay-raises-50-million-to-control-what-enterprise-ai-agents-can-do/ (read 12 Nov 2026) - www.calcalistech.com/ctechnews/article/bjx11v5x0gx: https://www.calcalistech.com/ctechnews/article/bjx11v5x0gx (read 12 Nov 2026) - thenextweb.com/news/bay-50m-ai-agent-security-stealth-battery-ventures: https://thenextweb.com/news/bay-50m-ai-agent-security-stealth-battery-ventures (read 13 Nov 2026) - www.knostic.ai/kirin: https://www.knostic.ai/kirin (read 12 Nov 2026) - www.knostic.ai/faq: https://www.knostic.ai/faq (read 12 Nov 2026) - www.knostic.ai/pricing: https://www.knostic.ai/pricing (read 12 Nov 2026) - www.knostic.ai/about: https://www.knostic.ai/about (read 12 Nov 2026) - www.knostic.ai/blog/when-an-mcp-tool-asks-for-your-github-token: https://www.knostic.ai/blog/when-an-mcp-tool-asks-for-your-github-token (read 13 Nov 2026) - docs.knostic.ai (checked, nothing found): https://docs.knostic.ai/ (read 12 Nov 2026) - news.google.com/rss/search?q=%22Knostic%22&hl=en-US&gl=US&ceid=US:en (checked, nothing found): https://news.google.com/rss/search?q=%22Knostic%22&hl=en-US&gl=US&ceid=US:en (read 13 Nov 2026) - www.securityweek.com/knostic-secures-11-million-to-rein-in-enterprise-ai-data-leakage-oversharing: https://www.securityweek.com/knostic-secures-11-million-to-rein-in-enterprise-ai-data-leakage-oversharing/ (read 12 Nov 2026) - finance.yahoo.com/sectors/technology/articles/knostic-named-2026-cb-insights-140000898.html: https://finance.yahoo.com/sectors/technology/articles/knostic-named-2026-cb-insights-140000898.html (read 12 Nov 2026) - www.knostic.ai/blog/gartner-cool-vendor-recognition: https://www.knostic.ai/blog/gartner-cool-vendor-recognition (read 12 Nov 2026) - www.knostic.ai (checked, nothing found): https://www.knostic.ai (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/compare/bay-vs-knostic/ --- # Bay beats WitnessAI 71.5 to 61.0 on the published weights HEAD TO HEAD, read 13 Nov 2026 **The short answer.** Bay scores 71.5 of 100 against WitnessAI at 61.0 on the published weights. Bay leads most on Deployment (9 to 5), Stops (8 to 6) and Beyond (7 to 6), while WitnessAI is ahead on Transparency (3 to 1). By situation, Bay leads in all five. ## Verdict on the published weights Bay scores 71.5 of 100 and WitnessAI scores 61.0. Bay is ahead on 4 of 7 criteria and behind on 1; 2 are level. Bay leads on Deployment (9 to 5), Stops (8 to 6), Beyond (7 to 6) and Proof (named customers, funding, analyst coverage) (8 to 7), while WitnessAI is ahead on Transparency (3 to 1). **Where Bay is behind.** Bay is a young company that left stealth in 2026, publishes no price and has little independent testing behind its claims. ## Criterion by criterion | Criterion | Weight | Bay | WitnessAI | | --- | --- | --- | --- | | Sees agentic activity on the endpoint | 25 | 8 | 8 | | Stops the action | 25 | 8 | 6 | | Deployment friction | 15 | 9 | 5 | | Beyond the endpoint | 10 | 7 | 6 | | Policy and governance | 10 | 6 | 6 | | Transparency | 10 | 1 | 3 | | Proof | 5 | 8 | 7 | | Total (0 to 100) | 100 | 71.5 | 61.0 | Bay against WitnessAI, scores out of 10 on the site weights. ## What each sees and can stop | Tool | Surface strip | Stop strip | | --- | --- | --- | | [Bay](/platforms/bay/) | Surfaces seen: Desktop apps partial; IDE assistants documented; Command-line agents documented; MCP servers documented; Browser extensions documented; SaaS copilots partial | Actions it can stop: File read partial; Shell command documented; Network call documented; Credential access documented; Data paste partial; Package install partial | | [WitnessAI](/platforms/witnessai/) | Surfaces seen: Desktop apps documented; IDE assistants documented; Command-line agents partial; MCP servers documented; Browser extensions partial; SaaS copilots partial | Actions it can stop: File read none found; Shell command partial; Network call documented; Credential access partial; Data paste documented; Package install none found | Filled documented, half partial, open none found. Surface order: desktop apps, IDE assistants, command-line agents, MCP servers, browser extensions, SaaS copilots. Stop order: file read, shell command, network call, credential access, data paste, package install. | Where they differ | Bay | WitnessAI | | --- | --- | --- | | Desktop apps | partial | documented | | Command-line agents | documented | partial | | Browser extensions | documented | partial | | Stops: file read | partial | none found | | Stops: shell command | documented | partial | | Stops: credential access | documented | partial | | Stops: data paste | partial | documented | | Stops: package install | partial | none found | 8 of 12 cells differ between Bay and WitnessAI as documented on 13 Nov 2026. ## Who should pick what, by situation - **Coding agents** ([an engineering organization where every developer runs coding agents with shell access](/best-for/coding-agents/)): Bay leads WitnessAI 75.5 to 62.5, ahead on Stops (8 to 6), Deployment (9 to 5), Beyond (7 to 6) and Proof (named customers, funding, analyst coverage) (8 to 7), while WitnessAI is ahead on Transparency (3 to 1). - **Regulated data** ([a regulated company that must show which AI tools touch customer data](/best-for/regulated-data/)): Bay leads WitnessAI 70.5 to 64.5, ahead on Stops (8 to 6), Deployment (9 to 5), Beyond (7 to 6) and Proof (named customers, funding, analyst coverage) (8 to 7), while WitnessAI is ahead on Transparency (3 to 1). - **Lean team with EDR** ([a lean security team with EDR in place and no view of AI agents](/best-for/lean-team-edr/)): Bay leads WitnessAI 76.5 to 62.5, ahead on Deployment (9 to 5), Stops (8 to 6), Beyond (7 to 6) and Proof (named customers, funding, analyst coverage) (8 to 7), while WitnessAI is ahead on Transparency (3 to 1). - **Staff rollout** ([a company rolling out an AI assistant and browser agents to all staff](/best-for/staff-rollout/)): Bay leads WitnessAI 73.5 to 61.5, ahead on Deployment (9 to 5), Stops (8 to 6), Beyond (7 to 6) and Proof (named customers, funding, analyst coverage) (8 to 7), while WitnessAI is ahead on Transparency (3 to 1). - **Board in 30 days** ([a CISO asked by the board for an inventory of AI agents within 30 days](/best-for/board-30-days/)): Bay leads WitnessAI 77.5 to 64.0, ahead on Deployment (9 to 5), Stops (8 to 6), Beyond (7 to 6) and Proof (named customers, funding, analyst coverage) (8 to 7), while WitnessAI is ahead on Transparency (3 to 1). ## What would change this ranking Bay is 10.5 points clear of WitnessAI on the published weights. No single score change could close that gap. Scores move only when a sourced fact changes ([how](/method/)). ## Read the profiles - [Bay, ranked 2nd of 16](/platforms/bay/): Bay documents on-device, session-aware allow/ask/deny/steer enforcement with named action types, agentless MDM/EDR deployment in under 5 minutes (Bay's claim), and one policy across endpoint, browser and cloud. - [WitnessAI, ranked 11th of 16](/platforms/witnessai/): WitnessAI documents inline blocking of unapproved MCP servers and tool calls across IDEs, chat apps and custom agents, with audit records per blocked call and one policy for humans and agents. ## Sources - softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control: https://softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control (read 12 Nov 2026) - siliconangle.com/2026/11/10/ai-security-startup-bay-launches-with-50m-to-vet-agent-actions-before-they-run: https://siliconangle.com/2026/11/10/ai-security-startup-bay-launches-with-50m-to-vet-agent-actions-before-they-run/ (read 12 Nov 2026) - siliconangle.com/2026/01/13/witnessai-debuts-agentic-security-enterprises-deploy-autonomous-ai-agents: https://siliconangle.com/2026/01/13/witnessai-debuts-agentic-security-enterprises-deploy-autonomous-ai-agents/ (read 12 Nov 2026) - www.helpnetsecurity.com/2026/06/17/witnessai-agentic-control: https://www.helpnetsecurity.com/2026/06/17/witnessai-agentic-control/ (read 13 Nov 2026) - pulse2.com/witnessai-profile-rick-caccia-interview: https://pulse2.com/witnessai-profile-rick-caccia-interview/ (read 13 Nov 2026) - www.securityweek.com/witnessai-raises-58-million-for-ai-security-platform: https://www.securityweek.com/witnessai-raises-58-million-for-ai-security-platform/ (read 12 Nov 2026) - www.bay.io: https://www.bay.io (read 12 Nov 2026) - www.bay.io/platform/ai-endpoint-security: https://www.bay.io/platform/ai-endpoint-security (read 12 Nov 2026) - www.securityweek.com/bay-raises-50-million-to-control-what-enterprise-ai-agents-can-do: https://www.securityweek.com/bay-raises-50-million-to-control-what-enterprise-ai-agents-can-do/ (read 12 Nov 2026) - thenextweb.com/news/bay-50m-ai-agent-security-stealth-battery-ventures: https://thenextweb.com/news/bay-50m-ai-agent-security-stealth-battery-ventures (read 13 Nov 2026) - witness.ai/product: https://witness.ai/product/ (read 12 Nov 2026) - witness.ai/resources/witnessai-announces-integration-with-f5-big-ip-to-deliver-visibility-and-control-over-encrypted-genai-activity: https://witness.ai/resources/witnessai-announces-integration-with-f5-big-ip-to-deliver-visibility-and-control-over-encrypted-genai-activity/ (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/compare/bay-vs-witnessai/ --- # Bay beats Pillar Security 71.5 to 53.0 on the published weights HEAD TO HEAD, read 13 Nov 2026 **The short answer.** Bay scores 71.5 of 100 against Pillar Security at 53.0 on the published weights. Bay leads most on Sees (8 to 6), Stops (8 to 6) and Beyond (7 to 2), while Pillar Security is ahead on Transparency (4 to 1). By situation, Bay leads in all five. ## Verdict on the published weights Bay scores 71.5 of 100 and Pillar Security scores 53.0. Bay is ahead on 6 of 7 criteria and behind on 1; 0 are level. Bay leads on Sees (8 to 6), Stops (8 to 6), Beyond (7 to 2), Deployment (9 to 6), Policy (6 to 5) and Proof (named customers, funding, analyst coverage) (8 to 6), while Pillar Security is ahead on Transparency (4 to 1). **Where Bay is behind.** Bay is a young company that left stealth in 2026, publishes no price and has little independent testing behind its claims. ## Criterion by criterion | Criterion | Weight | Bay | Pillar Security | | --- | --- | --- | --- | | Sees agentic activity on the endpoint | 25 | 8 | 6 | | Stops the action | 25 | 8 | 6 | | Deployment friction | 15 | 9 | 6 | | Beyond the endpoint | 10 | 7 | 2 | | Policy and governance | 10 | 6 | 5 | | Transparency | 10 | 1 | 4 | | Proof | 5 | 8 | 6 | | Total (0 to 100) | 100 | 71.5 | 53.0 | Bay against Pillar Security, scores out of 10 on the site weights. ## What each sees and can stop | Tool | Surface strip | Stop strip | | --- | --- | --- | | [Bay](/platforms/bay/) | Surfaces seen: Desktop apps partial; IDE assistants documented; Command-line agents documented; MCP servers documented; Browser extensions documented; SaaS copilots partial | Actions it can stop: File read partial; Shell command documented; Network call documented; Credential access documented; Data paste partial; Package install partial | | [Pillar Security](/platforms/pillar-security/) | Surfaces seen: Desktop apps none found; IDE assistants documented; Command-line agents documented; MCP servers documented; Browser extensions none found; SaaS copilots none found | Actions it can stop: File read documented; Shell command documented; Network call documented; Credential access documented; Data paste none found; Package install none found | Filled documented, half partial, open none found. Surface order: desktop apps, IDE assistants, command-line agents, MCP servers, browser extensions, SaaS copilots. Stop order: file read, shell command, network call, credential access, data paste, package install. | Where they differ | Bay | Pillar Security | | --- | --- | --- | | Desktop apps | partial | none found | | Browser extensions | documented | none found | | SaaS copilots | partial | none found | | Stops: file read | partial | documented | | Stops: data paste | partial | none found | | Stops: package install | partial | none found | 6 of 12 cells differ between Bay and Pillar Security as documented on 13 Nov 2026. ## Who should pick what, by situation - **Coding agents** ([an engineering organization where every developer runs coding agents with shell access](/best-for/coding-agents/)): Bay leads Pillar Security 75.5 to 56.0, ahead on Stops (8 to 6), Sees (8 to 6), Deployment (9 to 6), Beyond (7 to 2), Policy (6 to 5) and Proof (named customers, funding, analyst coverage) (8 to 6), while Pillar Security is ahead on Transparency (4 to 1). - **Regulated data** ([a regulated company that must show which AI tools touch customer data](/best-for/regulated-data/)): Bay leads Pillar Security 70.5 to 50.5, ahead on Beyond (7 to 2), Sees (8 to 6), Stops (8 to 6), Policy (6 to 5), Deployment (9 to 6) and Proof (named customers, funding, analyst coverage) (8 to 6), while Pillar Security is ahead on Transparency (4 to 1). - **Lean team with EDR** ([a lean security team with EDR in place and no view of AI agents](/best-for/lean-team-edr/)): Bay leads Pillar Security 76.5 to 56.0, ahead on Deployment (9 to 6), Sees (8 to 6), Stops (8 to 6), Beyond (7 to 2), Policy (6 to 5) and Proof (named customers, funding, analyst coverage) (8 to 6), while Pillar Security is ahead on Transparency (4 to 1). - **Staff rollout** ([a company rolling out an AI assistant and browser agents to all staff](/best-for/staff-rollout/)): Bay leads Pillar Security 73.5 to 48.0, ahead on Beyond (7 to 2), Deployment (9 to 6), Sees (8 to 6), Stops (8 to 6), Policy (6 to 5) and Proof (named customers, funding, analyst coverage) (8 to 6), while Pillar Security is ahead on Transparency (4 to 1). - **Board in 30 days** ([a CISO asked by the board for an inventory of AI agents within 30 days](/best-for/board-30-days/)): Bay leads Pillar Security 77.5 to 54.5, ahead on Deployment (9 to 6), Sees (8 to 6), Beyond (7 to 2), Stops (8 to 6), Proof (named customers, funding, analyst coverage) (8 to 6) and Policy (6 to 5), while Pillar Security is ahead on Transparency (4 to 1). ## What would change this ranking Bay is 18.5 points clear of Pillar Security on the published weights. No single score change could close that gap. Scores move only when a sourced fact changes ([how](/method/)). ## Read the profiles - [Bay, ranked 2nd of 16](/platforms/bay/): Bay documents on-device, session-aware allow/ask/deny/steer enforcement with named action types, agentless MDM/EDR deployment in under 5 minutes (Bay's claim), and one policy across endpoint, browser and cloud. - [Pillar Security, ranked 14th of 16](/platforms/pillar-security/): Documents pre-action blocking in the agent's own pre-tool hook for Claude Code, Codex CLI and Cursor, with credential-exfiltration and persistence controls and MDM-installed hooks developers cannot disable. ## Sources - softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control: https://softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control (read 12 Nov 2026) - siliconangle.com/2026/11/10/ai-security-startup-bay-launches-with-50m-to-vet-agent-actions-before-they-run: https://siliconangle.com/2026/11/10/ai-security-startup-bay-launches-with-50m-to-vet-agent-actions-before-they-run/ (read 12 Nov 2026) - thehackernews.com/2026/04/google-fixes-cvss-10-gemini-cli-ci-rce.html: https://thehackernews.com/2026/04/google-fixes-cvss-10-gemini-cli-ci-rce.html (read 12 Nov 2026) - thehackernews.com/2026/10/opencode-flaw-lets-malicious-mcp.html: https://thehackernews.com/2026/10/opencode-flaw-lets-malicious-mcp.html (read 12 Nov 2026) - www.infosecurity-magazine.com/news/two-critical-flaws-in-n8n-ai: https://www.infosecurity-magazine.com/news/two-critical-flaws-in-n8n-ai/ (read 12 Nov 2026) - docs.pillar.security: https://docs.pillar.security (read 12 Nov 2026) - www.calcalistech.com/ctechnews/article/hkjzjkjrfl: https://www.calcalistech.com/ctechnews/article/hkjzjkjrfl (read 12 Nov 2026) - www.bay.io: https://www.bay.io (read 12 Nov 2026) - www.bay.io/platform/ai-endpoint-security: https://www.bay.io/platform/ai-endpoint-security (read 12 Nov 2026) - www.securityweek.com/bay-raises-50-million-to-control-what-enterprise-ai-agents-can-do: https://www.securityweek.com/bay-raises-50-million-to-control-what-enterprise-ai-agents-can-do/ (read 12 Nov 2026) - thenextweb.com/news/bay-50m-ai-agent-security-stealth-battery-ventures: https://thenextweb.com/news/bay-50m-ai-agent-security-stealth-battery-ventures (read 13 Nov 2026) - www.bay.io/platform/cloud-workspaces: https://www.bay.io/platform/cloud-workspaces (read 12 Nov 2026) - www.bay.io/platform/browser: https://www.bay.io/platform/browser (read 12 Nov 2026) - www.bay.io/blog/bay-emerges-from-stealth-with-50m-after-fortune-100-companies-used-its-platform-to-triple-autonomous-ai-agent-use-2: https://www.bay.io/blog/bay-emerges-from-stealth-with-50m-after-fortune-100-companies-used-its-platform-to-triple-autonomous-ai-agent-use-2 (read 12 Nov 2026) - pulse2.com/bay-50-million-funding-to-expand-ai-agent-security-platform-across-enterprise-environments: https://pulse2.com/bay-50-million-funding-to-expand-ai-agent-security-platform-across-enterprise-environments/ (read 13 Nov 2026) - www.bay.io/blog/introducing-bay-the-ai-agent-security-platform-out-of-stealth-with-50m: https://www.bay.io/blog/introducing-bay-the-ai-agent-security-platform-out-of-stealth-with-50m (read 12 Nov 2026) - docs.bay.io: https://docs.bay.io (read 12 Nov 2026) - www.bay.io/sitemap.xml (checked, nothing found): https://www.bay.io/sitemap.xml (read 12 Nov 2026) - www.bay.io/blog (checked, nothing found): https://www.bay.io/blog (read 12 Nov 2026) - www.pillar.security/platform/agentic-endpoint: https://www.pillar.security/platform/agentic-endpoint (read 12 Nov 2026) - news.google.com/rss/search?q=%22Pillar+Security%22&hl=en-US&gl=US&ceid=US:en (checked, nothing found): https://news.google.com/rss/search?q=%22Pillar+Security%22&hl=en-US&gl=US&ceid=US:en (read 13 Nov 2026) - www.pillar.security/blog/introducing-pillar-for-ai-coding-agents: https://www.pillar.security/blog/introducing-pillar-for-ai-coding-agents (read 12 Nov 2026) - www.pillar.security/blog/inspecting-the-inspector-rce-in-mcps-official-debugging-tool: https://www.pillar.security/blog/inspecting-the-inspector-rce-in-mcps-official-debugging-tool (read 13 Nov 2026) - www.pillar.security/pricing (checked, nothing found): https://www.pillar.security/pricing (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/compare/bay-vs-pillar-security/ --- # Bay beats Aim Security 71.5 to 69.0 on the published weights HEAD TO HEAD, read 13 Nov 2026 **The short answer.** Bay scores 71.5 of 100 against Aim Security at 69.0 on the published weights. Bay leads on Deployment (9 to 5) and Stops (8 to 7), while Aim Security is ahead on Transparency (7 to 1). By situation, Bay leads in all five. Bay is ahead on 2 of 7 criteria. ## Verdict on the published weights Bay scores 71.5 of 100 and Aim Security scores 69.0. Bay is ahead on 2 of 7 criteria and behind on 1; 4 are level. Bay leads on Deployment (9 to 5) and Stops (8 to 7), while Aim Security is ahead on Transparency (7 to 1). **Where Bay is behind.** Bay is a young company that left stealth in 2026, publishes no price and has little independent testing behind its claims. ## Criterion by criterion | Criterion | Weight | Bay | Aim Security | | --- | --- | --- | --- | | Sees agentic activity on the endpoint | 25 | 8 | 8 | | Stops the action | 25 | 8 | 7 | | Deployment friction | 15 | 9 | 5 | | Beyond the endpoint | 10 | 7 | 7 | | Policy and governance | 10 | 6 | 6 | | Transparency | 10 | 1 | 7 | | Proof | 5 | 8 | 8 | | Total (0 to 100) | 100 | 71.5 | 69.0 | Bay against Aim Security, scores out of 10 on the site weights. ## What each sees and can stop | Tool | Surface strip | Stop strip | | --- | --- | --- | | [Bay](/platforms/bay/) | Surfaces seen: Desktop apps partial; IDE assistants documented; Command-line agents documented; MCP servers documented; Browser extensions documented; SaaS copilots partial | Actions it can stop: File read partial; Shell command documented; Network call documented; Credential access documented; Data paste partial; Package install partial | | [Aim Security](/platforms/aim-security/) | Surfaces seen: Desktop apps documented; IDE assistants partial; Command-line agents documented; MCP servers documented; Browser extensions documented; SaaS copilots partial | Actions it can stop: File read partial; Shell command partial; Network call partial; Credential access none found; Data paste documented; Package install none found | Filled documented, half partial, open none found. Surface order: desktop apps, IDE assistants, command-line agents, MCP servers, browser extensions, SaaS copilots. Stop order: file read, shell command, network call, credential access, data paste, package install. | Where they differ | Bay | Aim Security | | --- | --- | --- | | Desktop apps | partial | documented | | IDE assistants | documented | partial | | Stops: shell command | documented | partial | | Stops: network call | documented | partial | | Stops: credential access | documented | none found | | Stops: data paste | partial | documented | | Stops: package install | partial | none found | 7 of 12 cells differ between Bay and Aim Security as documented on 13 Nov 2026. ## Who should pick what, by situation - **Coding agents** ([an engineering organization where every developer runs coding agents with shell access](/best-for/coding-agents/)): Bay leads Aim Security 75.5 to 69.0, ahead on Deployment (9 to 5) and Stops (8 to 7), while Aim Security is ahead on Transparency (7 to 1). - **Regulated data** ([a regulated company that must show which AI tools touch customer data](/best-for/regulated-data/)): Bay leads Aim Security 70.5 to 70.0, ahead on Deployment (9 to 5) and Stops (8 to 7), while Aim Security is ahead on Transparency (7 to 1). - **Lean team with EDR** ([a lean security team with EDR in place and no view of AI agents](/best-for/lean-team-edr/)): Bay leads Aim Security 76.5 to 67.5, ahead on Deployment (9 to 5) and Stops (8 to 7), while Aim Security is ahead on Transparency (7 to 1). - **Staff rollout** ([a company rolling out an AI assistant and browser agents to all staff](/best-for/staff-rollout/)): Bay leads Aim Security 73.5 to 68.5, ahead on Deployment (9 to 5) and Stops (8 to 7), while Aim Security is ahead on Transparency (7 to 1). - **Board in 30 days** ([a CISO asked by the board for an inventory of AI agents within 30 days](/best-for/board-30-days/)): Bay leads Aim Security 77.5 to 68.0, ahead on Deployment (9 to 5) and Stops (8 to 7), while Aim Security is ahead on Transparency (7 to 1). ## What would change this ranking Bay is 2.5 points clear of Aim Security on the published weights. The smallest single change that would put Aim Security first is a Deployment score of 7 instead of 5 (weight 15). Scores move only when a sourced fact changes ([how](/method/)). ## Read the profiles - [Bay, ranked 2nd of 16](/platforms/bay/): Bay documents on-device, session-aware allow/ask/deny/steer enforcement with named action types, agentless MDM/EDR deployment in under 5 minutes (Bay's claim), and one policy across endpoint, browser and cloud. - [Aim Security, ranked 4th of 16](/platforms/aim-security/): Cato AI Endpoint documents a lightweight client that finds local agents with their MCP servers and skills and blocks an unapproved tool call before it runs, in one dashboard with inline network inspection. ## Sources - softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control: https://softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control (read 12 Nov 2026) - siliconangle.com/2026/11/10/ai-security-startup-bay-launches-with-50m-to-vet-agent-actions-before-they-run: https://siliconangle.com/2026/11/10/ai-security-startup-bay-launches-with-50m-to-vet-agent-actions-before-they-run/ (read 12 Nov 2026) - knowledge.catonetworks.com/docs/product-updates-october-19-2026.md: https://knowledge.catonetworks.com/docs/product-updates-october-19-2026.md (read 13 Nov 2026) - knowledge.catonetworks.com/docs/product-updates-november-9-2026.md: https://knowledge.catonetworks.com/docs/product-updates-november-9-2026.md (read 13 Nov 2026) - knowledge.catonetworks.com/docs/integrating-anthropic-compliance-api-with-cato-ai-security: https://knowledge.catonetworks.com/docs/integrating-anthropic-compliance-api-with-cato-ai-security (read 12 Nov 2026) - techcrunch.com/2024/01/31/aim-security-raises-10m-for-its-genai-security-platform: https://techcrunch.com/2024/01/31/aim-security-raises-10m-for-its-genai-security-platform/ (read 12 Nov 2026) - candid.technology/best-shadow-agent-monitoring-solutions-for-enterprises: https://candid.technology/best-shadow-agent-monitoring-solutions-for-enterprises/ (read 13 Nov 2026) - www.bay.io: https://www.bay.io (read 12 Nov 2026) - www.bay.io/platform/ai-endpoint-security: https://www.bay.io/platform/ai-endpoint-security (read 12 Nov 2026) - www.securityweek.com/bay-raises-50-million-to-control-what-enterprise-ai-agents-can-do: https://www.securityweek.com/bay-raises-50-million-to-control-what-enterprise-ai-agents-can-do/ (read 12 Nov 2026) - thenextweb.com/news/bay-50m-ai-agent-security-stealth-battery-ventures: https://thenextweb.com/news/bay-50m-ai-agent-security-stealth-battery-ventures (read 13 Nov 2026) - www.bay.io/blog/introducing-bay-the-ai-agent-security-platform-out-of-stealth-with-50m: https://www.bay.io/blog/introducing-bay-the-ai-agent-security-platform-out-of-stealth-with-50m (read 12 Nov 2026) - docs.bay.io: https://docs.bay.io (read 12 Nov 2026) - www.bay.io/sitemap.xml (checked, nothing found): https://www.bay.io/sitemap.xml (read 12 Nov 2026) - www.bay.io/blog (checked, nothing found): https://www.bay.io/blog (read 12 Nov 2026) - www.catonetworks.com/platform/cato-ai-endpoint: https://www.catonetworks.com/platform/cato-ai-endpoint/ (read 12 Nov 2026) - knowledge.catonetworks.com/docs/product-updates-october-12-2026.md: https://knowledge.catonetworks.com/docs/product-updates-october-12-2026.md (read 13 Nov 2026) - knowledge.catonetworks.com/llms.txt (checked, nothing found): https://knowledge.catonetworks.com/llms.txt (read 12 Nov 2026) - knowledge.catonetworks.com/docs/2026-release-notes: https://knowledge.catonetworks.com/docs/2026-release-notes (read 12 Nov 2026) - services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2025-32711: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2025-32711 (read 12 Nov 2026) - thehackernews.com/2025/06/zero-click-ai-vulnerability-exposes.html: https://thehackernews.com/2025/06/zero-click-ai-vulnerability-exposes.html (read 12 Nov 2026) - www.securityweek.com/echoleak-ai-attack-enabled-theft-of-sensitive-data-via-microsoft-365-copilot: https://www.securityweek.com/echoleak-ai-attack-enabled-theft-of-sensitive-data-via-microsoft-365-copilot/ (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/compare/bay-vs-aim-security/ --- # Island beats LayerX 66.0 to 50.0 on the published weights HEAD TO HEAD, read 13 Nov 2026 **The short answer.** Island scores 66.0 of 100 against LayerX at 50.0 on the published weights. Island leads most on Sees (7 to 5), Stops (7 to 5) and Beyond (8 to 5), while LayerX is ahead on Transparency (4 to 3). By situation, Island leads in all five. ## Verdict on the published weights Island scores 66.0 of 100 and LayerX scores 50.0. Island is ahead on 6 of 7 criteria and behind on 1; 0 are level. Island leads on Sees (7 to 5), Stops (7 to 5), Beyond (8 to 5), Deployment (6 to 5), Proof (named customers, funding, analyst coverage) (8 to 5) and Policy (7 to 6), while LayerX is ahead on Transparency (4 to 3). ## Criterion by criterion | Criterion | Weight | Island | LayerX | | --- | --- | --- | --- | | Sees agentic activity on the endpoint | 25 | 7 | 5 | | Stops the action | 25 | 7 | 5 | | Deployment friction | 15 | 6 | 5 | | Beyond the endpoint | 10 | 8 | 5 | | Policy and governance | 10 | 7 | 6 | | Transparency | 10 | 3 | 4 | | Proof | 5 | 8 | 5 | | Total (0 to 100) | 100 | 66.0 | 50.0 | Island against LayerX, scores out of 10 on the site weights. ## What each sees and can stop | Tool | Surface strip | Stop strip | | --- | --- | --- | | [Island](/platforms/island/) | Surfaces seen: Desktop apps documented; IDE assistants documented; Command-line agents partial; MCP servers documented; Browser extensions documented; SaaS copilots partial | Actions it can stop: File read documented; Shell command documented; Network call documented; Credential access partial; Data paste documented; Package install partial | | [LayerX](/platforms/layerx/) | Surfaces seen: Desktop apps documented; IDE assistants none found; Command-line agents none found; MCP servers none found; Browser extensions documented; SaaS copilots partial | Actions it can stop: File read none found; Shell command none found; Network call none found; Credential access none found; Data paste documented; Package install partial | Filled documented, half partial, open none found. Surface order: desktop apps, IDE assistants, command-line agents, MCP servers, browser extensions, SaaS copilots. Stop order: file read, shell command, network call, credential access, data paste, package install. | Where they differ | Island | LayerX | | --- | --- | --- | | IDE assistants | documented | none found | | Command-line agents | partial | none found | | MCP servers | documented | none found | | Stops: file read | documented | none found | | Stops: shell command | documented | none found | | Stops: network call | documented | none found | | Stops: credential access | partial | none found | 7 of 12 cells differ between Island and LayerX as documented on 13 Nov 2026. ## Who should pick what, by situation - **Coding agents** ([an engineering organization where every developer runs coding agents with shell access](/best-for/coding-agents/)): Island leads LayerX 67.5 to 50.5, ahead on Stops (7 to 5), Sees (7 to 5), Deployment (6 to 5), Beyond (8 to 5), Proof (named customers, funding, analyst coverage) (8 to 5) and Policy (7 to 6), while LayerX is ahead on Transparency (4 to 3). - **Regulated data** ([a regulated company that must show which AI tools touch customer data](/best-for/regulated-data/)): Island leads LayerX 69.5 to 52.0, ahead on Sees (7 to 5), Beyond (8 to 5), Stops (7 to 5), Policy (7 to 6), Proof (named customers, funding, analyst coverage) (8 to 5) and Deployment (6 to 5), while LayerX is ahead on Transparency (4 to 3). - **Lean team with EDR** ([a lean security team with EDR in place and no view of AI agents](/best-for/lean-team-edr/)): Island leads LayerX 66.5 to 50.5, ahead on Sees (7 to 5), Stops (7 to 5), Deployment (6 to 5), Beyond (8 to 5), Proof (named customers, funding, analyst coverage) (8 to 5) and Policy (7 to 6), while LayerX is ahead on Transparency (4 to 3). - **Staff rollout** ([a company rolling out an AI assistant and browser agents to all staff](/best-for/staff-rollout/)): Island leads LayerX 69.5 to 50.5, ahead on Beyond (8 to 5), Sees (7 to 5), Stops (7 to 5), Deployment (6 to 5), Proof (named customers, funding, analyst coverage) (8 to 5) and Policy (7 to 6), while LayerX is ahead on Transparency (4 to 3). - **Board in 30 days** ([a CISO asked by the board for an inventory of AI agents within 30 days](/best-for/board-30-days/)): Island leads LayerX 66.5 to 50.0, ahead on Sees (7 to 5), Deployment (6 to 5), Beyond (8 to 5), Proof (named customers, funding, analyst coverage) (8 to 5), Stops (7 to 5) and Policy (7 to 6), while LayerX is ahead on Transparency (4 to 3). ## What would change this ranking Island is 16.0 points clear of LayerX on the published weights. No single score change could close that gap. Scores move only when a sourced fact changes ([how](/method/)). ## Read the profiles - [Island, ranked 7th of 16](/platforms/island/): Island combines an endpoint agent for Windows and macOS with a browser, extension and MCP gateway, and documents blocking of Bash execution, file reads and network connections as well as clipboard DLP. - [LayerX, ranked 15th of 16](/platforms/layerx/): It enforces warn, redact, block or guide policy on prompts, uploads and copy-paste at the moment of interaction with no proxy or browser replacement, and covers agentic browsers. ## Sources - www.securityweek.com/island-raises-400-million-at-6-4-billion-valuation: https://www.securityweek.com/island-raises-400-million-at-6-4-billion-valuation/ (read 12 Nov 2026) - securitybrief.co.uk/story/island-named-only-gartner-customers-choice-for-browsers: https://securitybrief.co.uk/story/island-named-only-gartner-customers-choice-for-browsers (read 13 Nov 2026) - securitybrief.co.uk/story/island-launches-enterprise-browser-to-control-ai-agents: https://securitybrief.co.uk/story/island-launches-enterprise-browser-to-control-ai-agents (read 13 Nov 2026) - thehackernews.com/2023/11/hands-on-review-layerxs-enterprise.html: https://thehackernews.com/2023/11/hands-on-review-layerxs-enterprise.html (read 12 Nov 2026) - thehackernews.com/2025/10/new-chatgpt-atlas-browser-exploit-lets.html: https://thehackernews.com/2025/10/new-chatgpt-atlas-browser-exploit-lets.html (read 12 Nov 2026) - www.infosecurity-magazine.com/news/bioshocking-ai-browser-prompt: https://www.infosecurity-magazine.com/news/bioshocking-ai-browser-prompt/ (read 13 Nov 2026) - www.island.io/product/enterprise-browser-extension: https://www.island.io/product/enterprise-browser-extension (read 12 Nov 2026) - www.island.io/press/island-brings-the-enterprise-browser-to-ai-agents: https://www.island.io/press/island-brings-the-enterprise-browser-to-ai-agents (read 12 Nov 2026) - www.island.io/product/ai: https://www.island.io/product/ai (read 12 Nov 2026) - www.island.io/blog/island-integrates-with-the-claude-compliance-api: https://www.island.io/blog/island-integrates-with-the-claude-compliance-api (read 12 Nov 2026) - www.island.io/blog/whats-new-in-island-ai-threat-intelligence: https://www.island.io/blog/whats-new-in-island-ai-threat-intelligence (read 13 Nov 2026) - www.securityweek.com/island-banks-250m-series-e-for-enterprise-browser: https://www.securityweek.com/island-banks-250m-series-e-for-enterprise-browser/ (read 12 Nov 2026) - www.island.io/about: https://www.island.io/about (read 12 Nov 2026) - www.island.io/blog/taskus-runs-on-default-deny-it-still-said-yes-to-ai: https://www.island.io/blog/taskus-runs-on-default-deny-it-still-said-yes-to-ai (read 12 Nov 2026) - www.island.io/press: https://www.island.io/press (read 12 Nov 2026) - www.akamai.com/products/akamai-workforce-protector: https://www.akamai.com/products/akamai-workforce-protector (read 12 Nov 2026) - www.akamai.com/newsroom/press-release/akamai-technologies-announces-intent-to-acquire-layerx-advancing-its-workforce-security-strategy-with-ai-usage-control: https://www.akamai.com/newsroom/press-release/akamai-technologies-announces-intent-to-acquire-layerx-advancing-its-workforce-security-strategy-with-ai-usage-control (read 12 Nov 2026) - news.google.com/rss/search?q=%22LayerX%22+browser+OR+SaaS+OR+copilot (checked, nothing found): https://news.google.com/rss/search?q=%22LayerX%22+browser+OR+SaaS+OR+copilot (read 13 Nov 2026) - www.securityweek.com/layerx-raises-26-million-for-browser-security-platform: https://www.securityweek.com/layerx-raises-26-million-for-browser-security-platform/ (read 12 Nov 2026) - www.securityweek.com/layerx-raises-75m-seed-funding-tackle-secure-web-browsing: https://www.securityweek.com/layerx-raises-75m-seed-funding-tackle-secure-web-browsing/ (read 12 Nov 2026) - www.akamai.com/blog/security/layerx-joins-akamai-to-scale-up-ai-usage-control: https://www.akamai.com/blog/security/layerx-joins-akamai-to-scale-up-ai-usage-control (read 12 Nov 2026) - www.akamai.com/blog/ai/ai-usage-control-employee-ai-use-needs-own-category: https://www.akamai.com/blog/ai/ai-usage-control-employee-ai-use-needs-own-category (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/compare/island-vs-layerx/ --- # Island beats Palo Alto Networks 66.0 to 45.0 on the published weights HEAD TO HEAD, read 13 Nov 2026 **The short answer.** Island scores 66.0 of 100 against Palo Alto Networks at 45.0 on the published weights. Island leads most on Sees (7 to 3), Stops (7 to 5) and Deployment (6 to 4), while Palo Alto Networks is ahead on Transparency (4 to 3). By situation, Island leads in all five. ## Verdict on the published weights Island scores 66.0 of 100 and Palo Alto Networks scores 45.0. Island is ahead on 5 of 7 criteria and behind on 1; 1 is level. Island leads on Sees (7 to 3), Stops (7 to 5), Deployment (6 to 4), Beyond (8 to 6) and Policy (7 to 5), while Palo Alto Networks is ahead on Transparency (4 to 3). ## Criterion by criterion | Criterion | Weight | Island | Palo Alto Networks | | --- | --- | --- | --- | | Sees agentic activity on the endpoint | 25 | 7 | 3 | | Stops the action | 25 | 7 | 5 | | Deployment friction | 15 | 6 | 4 | | Beyond the endpoint | 10 | 8 | 6 | | Policy and governance | 10 | 7 | 5 | | Transparency | 10 | 3 | 4 | | Proof | 5 | 8 | 8 | | Total (0 to 100) | 100 | 66.0 | 45.0 | Island against Palo Alto Networks, scores out of 10 on the site weights. ## What each sees and can stop | Tool | Surface strip | Stop strip | | --- | --- | --- | | [Island](/platforms/island/) | Surfaces seen: Desktop apps documented; IDE assistants documented; Command-line agents partial; MCP servers documented; Browser extensions documented; SaaS copilots partial | Actions it can stop: File read documented; Shell command documented; Network call documented; Credential access partial; Data paste documented; Package install partial | | [Palo Alto Networks](/platforms/palo-alto-networks/) | Surfaces seen: Desktop apps partial; IDE assistants documented; Command-line agents partial; MCP servers none found; Browser extensions documented; SaaS copilots partial | Actions it can stop: File read none found; Shell command partial; Network call partial; Credential access none found; Data paste documented; Package install none found | Filled documented, half partial, open none found. Surface order: desktop apps, IDE assistants, command-line agents, MCP servers, browser extensions, SaaS copilots. Stop order: file read, shell command, network call, credential access, data paste, package install. | Where they differ | Island | Palo Alto Networks | | --- | --- | --- | | Desktop apps | documented | partial | | MCP servers | documented | none found | | Stops: file read | documented | none found | | Stops: shell command | documented | partial | | Stops: network call | documented | partial | | Stops: credential access | partial | none found | | Stops: package install | partial | none found | 7 of 12 cells differ between Island and Palo Alto Networks as documented on 13 Nov 2026. ## Who should pick what, by situation - **Coding agents** ([an engineering organization where every developer runs coding agents with shell access](/best-for/coding-agents/)): Island leads Palo Alto Networks 67.5 to 45.0, ahead on Sees (7 to 3), Stops (7 to 5), Deployment (6 to 4), Policy (7 to 5) and Beyond (8 to 6), while Palo Alto Networks is ahead on Transparency (4 to 3). - **Regulated data** ([a regulated company that must show which AI tools touch customer data](/best-for/regulated-data/)): Island leads Palo Alto Networks 69.5 to 46.0, ahead on Sees (7 to 3), Policy (7 to 5), Stops (7 to 5), Beyond (8 to 6) and Deployment (6 to 4), while Palo Alto Networks is ahead on Transparency (4 to 3). - **Lean team with EDR** ([a lean security team with EDR in place and no view of AI agents](/best-for/lean-team-edr/)): Island leads Palo Alto Networks 66.5 to 43.0, ahead on Sees (7 to 3), Deployment (6 to 4), Stops (7 to 5), Policy (7 to 5) and Beyond (8 to 6), while Palo Alto Networks is ahead on Transparency (4 to 3). - **Staff rollout** ([a company rolling out an AI assistant and browser agents to all staff](/best-for/staff-rollout/)): Island leads Palo Alto Networks 69.5 to 48.0, ahead on Sees (7 to 3), Beyond (8 to 6), Stops (7 to 5), Deployment (6 to 4) and Policy (7 to 5), while Palo Alto Networks is ahead on Transparency (4 to 3). - **Board in 30 days** ([a CISO asked by the board for an inventory of AI agents within 30 days](/best-for/board-30-days/)): Island leads Palo Alto Networks 66.5 to 41.0, ahead on Sees (7 to 3), Deployment (6 to 4), Beyond (8 to 6), Stops (7 to 5) and Policy (7 to 5), while Palo Alto Networks is ahead on Transparency (4 to 3). ## What would change this ranking Island is 21.0 points clear of Palo Alto Networks on the published weights. No single score change could close that gap. Scores move only when a sourced fact changes ([how](/method/)). ## Read the profiles - [Island, ranked 7th of 16](/platforms/island/): Island combines an endpoint agent for Windows and macOS with a browser, extension and MCP gateway, and documents blocking of Bash execution, file reads and network connections as well as clipboard DLP. - [Palo Alto Networks, ranked 16th of 16](/platforms/palo-alto-networks/): Palo Alto Networks documents both a browser-level control point (extension inventory, copy-paste and upload limits for unsanctioned AI apps) and endpoint coding-agent policy enforcement under Prisma AIRS, backed by public docs and analyst recognition. ## Sources - www.securityweek.com/island-raises-400-million-at-6-4-billion-valuation: https://www.securityweek.com/island-raises-400-million-at-6-4-billion-valuation/ (read 12 Nov 2026) - securitybrief.co.uk/story/island-named-only-gartner-customers-choice-for-browsers: https://securitybrief.co.uk/story/island-named-only-gartner-customers-choice-for-browsers (read 13 Nov 2026) - securitybrief.co.uk/story/island-launches-enterprise-browser-to-control-ai-agents: https://securitybrief.co.uk/story/island-launches-enterprise-browser-to-control-ai-agents (read 13 Nov 2026) - docs.paloaltonetworks.com/ai-runtime-security/administration/agent-discovery/ai-agent-discovery-limitations: https://docs.paloaltonetworks.com/ai-runtime-security/administration/agent-discovery/ai-agent-discovery-limitations (read 13 Nov 2026) - www.helpnetsecurity.com/2026/09/29/palo-alto-networks-nvidia-ai-agent-security: https://www.helpnetsecurity.com/2026/09/29/palo-alto-networks-nvidia-ai-agent-security/ (read 13 Nov 2026) - siliconangle.com/2026/08/20/zero-networks-expands-palo-alto-networks-integration-to-ai-agent-control: https://siliconangle.com/2026/08/20/zero-networks-expands-palo-alto-networks-integration-to-ai-agent-control/ (read 13 Nov 2026) - www.island.io/product/ai: https://www.island.io/product/ai (read 12 Nov 2026) - www.island.io/blog/expanding-island-to-govern-your-ai-workforce: https://www.island.io/blog/expanding-island-to-govern-your-ai-workforce (read 12 Nov 2026) - www.island.io/blog/claude-code-cowork-and-claude-for-microsoft-365-now-report-into-island: https://www.island.io/blog/claude-code-cowork-and-claude-for-microsoft-365-now-report-into-island (read 12 Nov 2026) - www.island.io/blog/see-score-and-control-every-ai-asset-on-your-endpoints: https://www.island.io/blog/see-score-and-control-every-ai-asset-on-your-endpoints (read 12 Nov 2026) - www.paloaltonetworks.com/prisma/prisma-ai-runtime-security: https://www.paloaltonetworks.com/prisma/prisma-ai-runtime-security (read 12 Nov 2026) - www.paloaltonetworks.com/sase/prisma-browser: https://www.paloaltonetworks.com/sase/prisma-browser (read 12 Nov 2026) - news.google.com/rss/search?q=%22Palo+Alto+Networks%22+(inventory+OR+discovery+OR+MCP+OR+%22coding+agents%22) (checked, nothing found): https://news.google.com/rss/search?q=%22Palo+Alto+Networks%22+(inventory+OR+discovery+OR+MCP+OR+%22coding+agents%22) (read 13 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/compare/island-vs-palo-alto-networks/ --- # Microsoft beats CrowdStrike 75.0 to 66.0 on the published weights HEAD TO HEAD, read 13 Nov 2026 **The short answer.** Microsoft scores 75.0 of 100 against CrowdStrike at 66.0 on the published weights. Microsoft leads most on Transparency (7 to 3), Stops (8 to 7) and Beyond (8 to 6), while CrowdStrike is ahead on Proof (named customers, funding, analyst coverage) (7 to 4). By situation, Microsoft leads in all five. ## Verdict on the published weights Microsoft scores 75.0 of 100 and CrowdStrike scores 66.0. Microsoft is ahead on 4 of 7 criteria and behind on 1; 2 are level. Microsoft leads on Transparency (7 to 3), Stops (8 to 7), Beyond (8 to 6) and Policy (9 to 7), while CrowdStrike is ahead on Proof (named customers, funding, analyst coverage) (7 to 4). ## Criterion by criterion | Criterion | Weight | Microsoft | CrowdStrike | | --- | --- | --- | --- | | Sees agentic activity on the endpoint | 25 | 8 | 8 | | Stops the action | 25 | 8 | 7 | | Deployment friction | 15 | 6 | 6 | | Beyond the endpoint | 10 | 8 | 6 | | Policy and governance | 10 | 9 | 7 | | Transparency | 10 | 7 | 3 | | Proof | 5 | 4 | 7 | | Total (0 to 100) | 100 | 75.0 | 66.0 | Microsoft against CrowdStrike, scores out of 10 on the site weights. ## What each sees and can stop | Tool | Surface strip | Stop strip | | --- | --- | --- | | [Microsoft](/platforms/microsoft/) | Surfaces seen: Desktop apps documented; IDE assistants documented; Command-line agents documented; MCP servers documented; Browser extensions partial; SaaS copilots documented | Actions it can stop: File read partial; Shell command documented; Network call partial; Credential access partial; Data paste documented; Package install none found | | [CrowdStrike](/platforms/crowdstrike/) | Surfaces seen: Desktop apps documented; IDE assistants documented; Command-line agents documented; MCP servers documented; Browser extensions partial; SaaS copilots partial | Actions it can stop: File read partial; Shell command partial; Network call partial; Credential access partial; Data paste documented; Package install documented | Filled documented, half partial, open none found. Surface order: desktop apps, IDE assistants, command-line agents, MCP servers, browser extensions, SaaS copilots. Stop order: file read, shell command, network call, credential access, data paste, package install. | Where they differ | Microsoft | CrowdStrike | | --- | --- | --- | | SaaS copilots | documented | partial | | Stops: shell command | documented | partial | | Stops: package install | none found | documented | 3 of 12 cells differ between Microsoft and CrowdStrike as documented on 13 Nov 2026. ## Who should pick what, by situation - **Coding agents** ([an engineering organization where every developer runs coding agents with shell access](/best-for/coding-agents/)): Microsoft leads CrowdStrike 75.5 to 68.5, ahead on Stops (8 to 7), Policy (9 to 7), Transparency (7 to 3) and Beyond (8 to 6), while CrowdStrike is ahead on Proof (named customers, funding, analyst coverage) (7 to 4). - **Regulated data** ([a regulated company that must show which AI tools touch customer data](/best-for/regulated-data/)): Microsoft leads CrowdStrike 79.0 to 69.0, ahead on Policy (9 to 7), Beyond (8 to 6), Transparency (7 to 3) and Stops (8 to 7), while CrowdStrike is ahead on Proof (named customers, funding, analyst coverage) (7 to 4). - **Lean team with EDR** ([a lean security team with EDR in place and no view of AI agents](/best-for/lean-team-edr/)): Microsoft leads CrowdStrike 73.5 to 68.0, ahead on Stops (8 to 7), Policy (9 to 7), Transparency (7 to 3) and Beyond (8 to 6), while CrowdStrike is ahead on Proof (named customers, funding, analyst coverage) (7 to 4). - **Staff rollout** ([a company rolling out an AI assistant and browser agents to all staff](/best-for/staff-rollout/)): Microsoft leads CrowdStrike 75.5 to 66.0, ahead on Beyond (8 to 6), Stops (8 to 7), Policy (9 to 7) and Transparency (7 to 3), while CrowdStrike is ahead on Proof (named customers, funding, analyst coverage) (7 to 4). - **Board in 30 days** ([a CISO asked by the board for an inventory of AI agents within 30 days](/best-for/board-30-days/)): Microsoft leads CrowdStrike 72.0 to 68.0, ahead on Beyond (8 to 6), Transparency (7 to 3), Policy (9 to 7) and Stops (8 to 7), while CrowdStrike is ahead on Proof (named customers, funding, analyst coverage) (7 to 4). ## What would change this ranking Microsoft is 9.0 points clear of CrowdStrike on the published weights. No single score change could close that gap. Scores move only when a sourced fact changes ([how](/method/)). ## Read the profiles - [Microsoft, ranked 1st of 16](/platforms/microsoft/): Defender discovers dozens of local agents and MCP servers on Windows and macOS (GA) and can block prompt-injection-driven actions in Claude Code, Codex CLI and GitHub Copilot CLI on Windows, with Endpoint DLP blocking pastes and uploads to AI sites. - [CrowdStrike, ranked 5th of 16](/platforms/crowdstrike/): Falcon Guardian pairs OS-level sensor telemetry with agent-level hooks that can block actions before they run in Claude Code, Codex, GitHub Copilot, OpenCode and Cursor, plus package blocking in the same sensor. ## Sources - learn.microsoft.com/en-us/defender-endpoint/local-agent-discovery-overview: https://learn.microsoft.com/en-us/defender-endpoint/local-agent-discovery-overview (read 12 Nov 2026) - learn.microsoft.com/en-us/defender-endpoint/ai-agent-support-matrix: https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-support-matrix (read 12 Nov 2026) - learn.microsoft.com/en-us/defender-endpoint/whats-new-in-microsoft-defender-endpoint: https://learn.microsoft.com/en-us/defender-endpoint/whats-new-in-microsoft-defender-endpoint (read 12 Nov 2026) - venturebeat.com/technology/microsoft-takes-agent-365-out-of-preview-as-shadow-ai-becomes-an-enterprise-threat: https://venturebeat.com/technology/microsoft-takes-agent-365-out-of-preview-as-shadow-ai-becomes-an-enterprise-threat (read 12 Nov 2026) - www.msspalert.com/news/crowdstrike-brings-ai-security-to-the-endpoint-and-the-timing-makes-sense: https://www.msspalert.com/news/crowdstrike-brings-ai-security-to-the-endpoint-and-the-timing-makes-sense (read 12 Nov 2026) - 247wallst.com/investing/2026/11/04/crowdstrike-is-spending-495-million-to-police-ai-agents: https://247wallst.com/investing/2026/11/04/crowdstrike-is-spending-495-million-to-police-ai-agents/ (read 13 Nov 2026) - learn.microsoft.com/en-us/defender-endpoint/ai-agent-runtime-protection-overview: https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-runtime-protection-overview (read 12 Nov 2026) - www.microsoft.com/en-us/microsoft-365/enterprise/microsoft365-plans-and-pricing: https://www.microsoft.com/en-us/microsoft-365/enterprise/microsoft365-plans-and-pricing (read 12 Nov 2026) - www.microsoft.com/en-us/microsoft-agent-365: https://www.microsoft.com/en-us/microsoft-agent-365 (read 12 Nov 2026) - learn.microsoft.com/en-us/microsoft-agent-365/overview: https://learn.microsoft.com/en-us/microsoft-agent-365/overview (read 13 Nov 2026) - news.google.com/rss/search?q=%22Agent+365%22+price+OR+pricing+per+user&hl=en-US&gl=US&ceid=US:en (checked, nothing found): https://news.google.com/rss/search?q=%22Agent+365%22+price+OR+pricing+per+user&hl=en-US&gl=US&ceid=US:en (read 13 Nov 2026) - www.microsoft.com/en-us/security/blog/2026/05/01/microsoft-agent-365-now-generally-available-expands-capabilities-and-integrations: https://www.microsoft.com/en-us/security/blog/2026/05/01/microsoft-agent-365-now-generally-available-expands-capabilities-and-integrations/ (read 12 Nov 2026) - petri.com/microsoft-agent-365-discover-govern-ai-agents: https://petri.com/microsoft-agent-365-discover-govern-ai-agents/ (read 12 Nov 2026) - www.microsoft.com/en-us/security/business/analyst-reports (checked, nothing found): https://www.microsoft.com/en-us/security/business/analyst-reports (read 12 Nov 2026) - www.crowdstrike.com/en-us/blog/crowdstrike-shadow-AI-visibility-service: https://www.crowdstrike.com/en-us/blog/crowdstrike-shadow-AI-visibility-service/ (read 12 Nov 2026) - www.crowdstrike.com/en-us/pricing: https://www.crowdstrike.com/en-us/pricing/ (read 12 Nov 2026) - falcon.crowdstrike.com/documentation (checked, nothing found): https://falcon.crowdstrike.com/documentation/ (read 12 Nov 2026) - www.bankinfosecurity.com/crowdstrike-buys-pangea-for-260m-to-guard-enterprise-ai-use: https://www.bankinfosecurity.com/crowdstrike-buys-pangea-for-260m-to-guard-enterprise-ai-use (read 12 Nov 2026) - www.crowdstrike.com/en-us/platform/falcon-guardian-aidr: https://www.crowdstrike.com/en-us/platform/falcon-guardian-aidr/ (read 12 Nov 2026) - www.crowdstrike.com/en-us/resources/customer-stories/taxslayer: https://www.crowdstrike.com/en-us/resources/customer-stories/taxslayer/ (read 12 Nov 2026) - www.crowdstrike.com/en-us/resources/customer-stories/so-drama-entertainment: https://www.crowdstrike.com/en-us/resources/customer-stories/so-drama-entertainment/ (read 12 Nov 2026) - www.crowdstrike.com/en-us/press-releases/crowdstrike-introduces-ai-kill-switch-for-ai-safety: https://www.crowdstrike.com/en-us/press-releases/crowdstrike-introduces-ai-kill-switch-for-ai-safety/ (read 13 Nov 2026) - www.crowdstrike.com/en-us/blog/falcon-guardian-defines-next-generation-of-ai-security: https://www.crowdstrike.com/en-us/blog/falcon-guardian-defines-next-generation-of-ai-security/ (read 12 Nov 2026) - news.google.com/rss/search?q=%22CrowdStrike%22+(customer+OR+funding+OR+Gartner+OR+Forrester+OR+analyst)&hl=en-US&gl=US&ceid=US:en (checked, nothing found): https://news.google.com/rss/search?q=%22CrowdStrike%22+(customer+OR+funding+OR+Gartner+OR+Forrester+OR+analyst)&hl=en-US&gl=US&ceid=US:en (read 13 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/compare/microsoft-vs-crowdstrike/ --- # Microsoft beats SentinelOne 75.0 to 62.0 on the published weights HEAD TO HEAD, read 13 Nov 2026 **The short answer.** Microsoft scores 75.0 of 100 against SentinelOne at 62.0 on the published weights. Microsoft leads most on Policy (9 to 6), Transparency (7 to 4) and Sees (8 to 7), while SentinelOne is ahead on Proof (named customers, funding, analyst coverage) (5 to 4). By situation, Microsoft leads in all five. ## Verdict on the published weights Microsoft scores 75.0 of 100 and SentinelOne scores 62.0. Microsoft is ahead on 6 of 7 criteria and behind on 1; 0 are level. Microsoft leads on Policy (9 to 6), Transparency (7 to 4), Sees (8 to 7), Stops (8 to 7), Deployment (6 to 5) and Beyond (8 to 7), while SentinelOne is ahead on Proof (named customers, funding, analyst coverage) (5 to 4). ## Criterion by criterion | Criterion | Weight | Microsoft | SentinelOne | | --- | --- | --- | --- | | Sees agentic activity on the endpoint | 25 | 8 | 7 | | Stops the action | 25 | 8 | 7 | | Deployment friction | 15 | 6 | 5 | | Beyond the endpoint | 10 | 8 | 7 | | Policy and governance | 10 | 9 | 6 | | Transparency | 10 | 7 | 4 | | Proof | 5 | 4 | 5 | | Total (0 to 100) | 100 | 75.0 | 62.0 | Microsoft against SentinelOne, scores out of 10 on the site weights. ## What each sees and can stop | Tool | Surface strip | Stop strip | | --- | --- | --- | | [Microsoft](/platforms/microsoft/) | Surfaces seen: Desktop apps documented; IDE assistants documented; Command-line agents documented; MCP servers documented; Browser extensions partial; SaaS copilots documented | Actions it can stop: File read partial; Shell command documented; Network call partial; Credential access partial; Data paste documented; Package install none found | | [SentinelOne](/platforms/sentinelone/) | Surfaces seen: Desktop apps documented; IDE assistants documented; Command-line agents documented; MCP servers documented; Browser extensions documented; SaaS copilots partial | Actions it can stop: File read none found; Shell command partial; Network call partial; Credential access partial; Data paste documented; Package install none found | Filled documented, half partial, open none found. Surface order: desktop apps, IDE assistants, command-line agents, MCP servers, browser extensions, SaaS copilots. Stop order: file read, shell command, network call, credential access, data paste, package install. | Where they differ | Microsoft | SentinelOne | | --- | --- | --- | | Browser extensions | partial | documented | | SaaS copilots | documented | partial | | Stops: file read | partial | none found | | Stops: shell command | documented | partial | 4 of 12 cells differ between Microsoft and SentinelOne as documented on 13 Nov 2026. ## Who should pick what, by situation - **Coding agents** ([an engineering organization where every developer runs coding agents with shell access](/best-for/coding-agents/)): Microsoft leads SentinelOne 75.5 to 63.5, ahead on Stops (8 to 7), Policy (9 to 6), Sees (8 to 7), Deployment (6 to 5), Transparency (7 to 4) and Beyond (8 to 7), while SentinelOne is ahead on Proof (named customers, funding, analyst coverage) (5 to 4). - **Regulated data** ([a regulated company that must show which AI tools touch customer data](/best-for/regulated-data/)): Microsoft leads SentinelOne 79.0 to 64.0, ahead on Policy (9 to 6), Sees (8 to 7), Stops (8 to 7), Beyond (8 to 7), Transparency (7 to 4) and Deployment (6 to 5), while SentinelOne is ahead on Proof (named customers, funding, analyst coverage) (5 to 4). - **Lean team with EDR** ([a lean security team with EDR in place and no view of AI agents](/best-for/lean-team-edr/)): Microsoft leads SentinelOne 73.5 to 61.5, ahead on Sees (8 to 7), Policy (9 to 6), Deployment (6 to 5), Stops (8 to 7), Transparency (7 to 4) and Beyond (8 to 7), while SentinelOne is ahead on Proof (named customers, funding, analyst coverage) (5 to 4). - **Staff rollout** ([a company rolling out an AI assistant and browser agents to all staff](/best-for/staff-rollout/)): Microsoft leads SentinelOne 75.5 to 63.5, ahead on Policy (9 to 6), Beyond (8 to 7), Sees (8 to 7), Stops (8 to 7), Deployment (6 to 5) and Transparency (7 to 4), while SentinelOne is ahead on Proof (named customers, funding, analyst coverage) (5 to 4). - **Board in 30 days** ([a CISO asked by the board for an inventory of AI agents within 30 days](/best-for/board-30-days/)): Microsoft leads SentinelOne 72.0 to 61.0, ahead on Sees (8 to 7), Deployment (6 to 5), Policy (9 to 6), Transparency (7 to 4), Beyond (8 to 7) and Stops (8 to 7), while SentinelOne is ahead on Proof (named customers, funding, analyst coverage) (5 to 4). ## What would change this ranking Microsoft is 13.0 points clear of SentinelOne on the published weights. No single score change could close that gap. Scores move only when a sourced fact changes ([how](/method/)). ## Read the profiles - [Microsoft, ranked 1st of 16](/platforms/microsoft/): Defender discovers dozens of local agents and MCP servers on Windows and macOS (GA) and can block prompt-injection-driven actions in Claude Code, Codex CLI and GitHub Copilot CLI on Windows, with Endpoint DLP blocking pastes and uploads to AI sites. - [SentinelOne, ranked 10th of 16](/platforms/sentinelone/): SentinelOne documents an MCP gateway with allow, block, filter or redact actions and runtime tool-call blocking, plus a lightweight agent and extension that discover shadow AI across browsers, IDEs and terminal assistants. ## Sources - learn.microsoft.com/en-us/defender-endpoint/local-agent-discovery-overview: https://learn.microsoft.com/en-us/defender-endpoint/local-agent-discovery-overview (read 12 Nov 2026) - learn.microsoft.com/en-us/defender-endpoint/ai-agent-support-matrix: https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-support-matrix (read 12 Nov 2026) - learn.microsoft.com/en-us/defender-endpoint/whats-new-in-microsoft-defender-endpoint: https://learn.microsoft.com/en-us/defender-endpoint/whats-new-in-microsoft-defender-endpoint (read 12 Nov 2026) - venturebeat.com/technology/microsoft-takes-agent-365-out-of-preview-as-shadow-ai-becomes-an-enterprise-threat: https://venturebeat.com/technology/microsoft-takes-agent-365-out-of-preview-as-shadow-ai-becomes-an-enterprise-threat (read 12 Nov 2026) - siliconangle.com/2026/03/23/sentinelone-snyk-introduce-new-tools-securing-ai-agents: https://siliconangle.com/2026/03/23/sentinelone-snyk-introduce-new-tools-securing-ai-agents/ (read 12 Nov 2026) - chromewebstore.google.com/search/prompt%20security: https://chromewebstore.google.com/search/prompt%20security (read 12 Nov 2026) - www.crn.com/news/security/2026/sentinelone-s-chris-corde-on-why-traditional-edr-is-flying-blind-on-ai-agents: https://www.crn.com/news/security/2026/sentinelone-s-chris-corde-on-why-traditional-edr-is-flying-blind-on-ai-agents (read 13 Nov 2026) - github.com/prompt-security/clawsec: https://github.com/prompt-security/clawsec (read 12 Nov 2026) - github.com/prompt-security: https://github.com/prompt-security (read 12 Nov 2026) - learn.microsoft.com/en-us/defender-endpoint/ai-agent-runtime-protection-overview: https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-runtime-protection-overview (read 12 Nov 2026) - learn.microsoft.com/en-us/purview/dlp-learn-about-dlp: https://learn.microsoft.com/en-us/purview/dlp-learn-about-dlp (read 12 Nov 2026) - learn.microsoft.com/en-us/defender-endpoint/microsoft-defender-endpoint: https://learn.microsoft.com/en-us/defender-endpoint/microsoft-defender-endpoint (read 12 Nov 2026) - learn.microsoft.com/en-us/microsoft-agent-365/overview: https://learn.microsoft.com/en-us/microsoft-agent-365/overview (read 12 Nov 2026) - www.microsoft.com/en-us/microsoft-365/enterprise/microsoft365-plans-and-pricing: https://www.microsoft.com/en-us/microsoft-365/enterprise/microsoft365-plans-and-pricing (read 12 Nov 2026) - www.microsoft.com/en-us/microsoft-agent-365: https://www.microsoft.com/en-us/microsoft-agent-365 (read 12 Nov 2026) - news.google.com/rss/search?q=%22Agent+365%22+price+OR+pricing+per+user&hl=en-US&gl=US&ceid=US:en (checked, nothing found): https://news.google.com/rss/search?q=%22Agent+365%22+price+OR+pricing+per+user&hl=en-US&gl=US&ceid=US:en (read 13 Nov 2026) - www.sentinelone.com/blog/prompt-security-for-agentic-ai: https://www.sentinelone.com/blog/prompt-security-for-agentic-ai/ (read 12 Nov 2026) - www.sentinelone.com/blog/a-new-chapter-for-ai-and-cybersecurity-sentinelone-acquires-prompt-security: https://www.sentinelone.com/blog/a-new-chapter-for-ai-and-cybersecurity-sentinelone-acquires-prompt-security/ (read 12 Nov 2026) - www.sentinelone.com/press/sentinelone-launches-singularity-agentic-detection-and-response-to-stop-rogue-and-malicious-ai-agents-at-machine-speed: https://www.sentinelone.com/press/sentinelone-launches-singularity-agentic-detection-and-response-to-stop-rogue-and-malicious-ai-agents-at-machine-speed/ (read 12 Nov 2026) - www.sentinelone.com/press/sentinelone-expands-ai-security-coverage-to-claude-cowork: https://www.sentinelone.com/press/sentinelone-expands-ai-security-coverage-to-claude-cowork/ (read 12 Nov 2026) - www.sentinelone.com/platform/prompt-security-ai-usage-control: https://www.sentinelone.com/platform/prompt-security-ai-usage-control/ (read 13 Nov 2026) - www.sentinelone.com/customers/10xbanking: https://www.sentinelone.com/customers/10xbanking/ (read 13 Nov 2026) - www.sentinelone.com/platform/prompt-security-agentic-ai-security (checked, nothing found): https://www.sentinelone.com/platform/prompt-security-agentic-ai-security/ (read 12 Nov 2026) - prompt.security/blog/when-your-plugin-starts-picking-your-dependencies-marketplace-skills-and-dependency-hijack-in-claude-code: https://prompt.security/blog/when-your-plugin-starts-picking-your-dependencies-marketplace-skills-and-dependency-hijack-in-claude-code (read 12 Nov 2026) - prompt.security/blog: https://prompt.security/blog (read 12 Nov 2026) - prompt.security/blog/when-ai-agents-become-the-supply-chain-hidden-control-planes-in-agentic-systems: https://prompt.security/blog/when-ai-agents-become-the-supply-chain-hidden-control-planes-in-agentic-systems (read 13 Nov 2026) - prompt.security/pricing (checked, nothing found): https://prompt.security/pricing (read 12 Nov 2026) - docs.prompt.security (checked, nothing found): https://docs.prompt.security/ (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/compare/microsoft-vs-sentinelone/ --- # Harmonic Security beats Knostic 71.5 to 64.0 on the published weights HEAD TO HEAD, read 13 Nov 2026 **The short answer.** Harmonic Security scores 71.5 of 100 against Knostic at 64.0 on the published weights. Harmonic Security leads most on Beyond (7 to 3), Sees (8 to 7) and Policy (7 to 6), while Knostic is ahead on Transparency (7 to 6). By situation, Harmonic Security leads in all five. ## Verdict on the published weights Harmonic Security scores 71.5 of 100 and Knostic scores 64.0. Harmonic Security is ahead on 4 of 7 criteria and behind on 1; 2 are level. Harmonic Security leads on Beyond (7 to 3), Sees (8 to 7), Policy (7 to 6) and Proof (named customers, funding, analyst coverage) (7 to 5), while Knostic is ahead on Transparency (7 to 6). ## Criterion by criterion | Criterion | Weight | Knostic | Harmonic Security | | --- | --- | --- | --- | | Sees agentic activity on the endpoint | 25 | 7 | 8 | | Stops the action | 25 | 7 | 7 | | Deployment friction | 15 | 7 | 7 | | Beyond the endpoint | 10 | 3 | 7 | | Policy and governance | 10 | 6 | 7 | | Transparency | 10 | 7 | 6 | | Proof | 5 | 5 | 7 | | Total (0 to 100) | 100 | 64.0 | 71.5 | Knostic against Harmonic Security, scores out of 10 on the site weights. ## What each sees and can stop | Tool | Surface strip | Stop strip | | --- | --- | --- | | [Knostic](/platforms/knostic/) | Surfaces seen: Desktop apps documented; IDE assistants documented; Command-line agents documented; MCP servers documented; Browser extensions partial; SaaS copilots partial | Actions it can stop: File read documented; Shell command documented; Network call partial; Credential access documented; Data paste partial; Package install partial | | [Harmonic Security](/platforms/harmonic-security/) | Surfaces seen: Desktop apps documented; IDE assistants documented; Command-line agents documented; MCP servers documented; Browser extensions documented; SaaS copilots partial | Actions it can stop: File read none found; Shell command partial; Network call partial; Credential access partial; Data paste documented; Package install none found | Filled documented, half partial, open none found. Surface order: desktop apps, IDE assistants, command-line agents, MCP servers, browser extensions, SaaS copilots. Stop order: file read, shell command, network call, credential access, data paste, package install. | Where they differ | Knostic | Harmonic Security | | --- | --- | --- | | Browser extensions | partial | documented | | Stops: file read | documented | none found | | Stops: shell command | documented | partial | | Stops: credential access | documented | partial | | Stops: data paste | partial | documented | | Stops: package install | partial | none found | 6 of 12 cells differ between Knostic and Harmonic Security as documented on 13 Nov 2026. ## Who should pick what, by situation - **Coding agents** ([an engineering organization where every developer runs coding agents with shell access](/best-for/coding-agents/)): Harmonic Security leads Knostic 72.0 to 66.0, ahead on Sees (8 to 7), Beyond (7 to 3), Policy (7 to 6) and Proof (named customers, funding, analyst coverage) (7 to 5), while Knostic is ahead on Transparency (7 to 6). - **Regulated data** ([a regulated company that must show which AI tools touch customer data](/best-for/regulated-data/)): Harmonic Security leads Knostic 72.5 to 60.5, ahead on Beyond (7 to 3), Sees (8 to 7), Policy (7 to 6) and Proof (named customers, funding, analyst coverage) (7 to 5), while Knostic is ahead on Transparency (7 to 6). - **Lean team with EDR** ([a lean security team with EDR in place and no view of AI agents](/best-for/lean-team-edr/)): Harmonic Security leads Knostic 72.5 to 66.0, ahead on Sees (8 to 7), Beyond (7 to 3), Policy (7 to 6) and Proof (named customers, funding, analyst coverage) (7 to 5), while Knostic is ahead on Transparency (7 to 6). - **Staff rollout** ([a company rolling out an AI assistant and browser agents to all staff](/best-for/staff-rollout/)): Harmonic Security leads Knostic 71.5 to 58.0, ahead on Beyond (7 to 3), Sees (8 to 7), Policy (7 to 6) and Proof (named customers, funding, analyst coverage) (7 to 5), while Knostic is ahead on Transparency (7 to 6). - **Board in 30 days** ([a CISO asked by the board for an inventory of AI agents within 30 days](/best-for/board-30-days/)): Harmonic Security leads Knostic 73.5 to 64.5, ahead on Sees (8 to 7), Beyond (7 to 3), Proof (named customers, funding, analyst coverage) (7 to 5) and Policy (7 to 6), while Knostic is ahead on Transparency (7 to 6). ## What would change this ranking Harmonic Security is 7.5 points clear of Knostic on the published weights. No single score change could close that gap. Scores move only when a sourced fact changes ([how](/method/)). ## Read the profiles - [Knostic, ranked 9th of 16](/platforms/knostic/): Knostic documents Enforce-mode blocking of secret file reads, risky MCP servers, destructive commands and unsafe packages inside agents, with a published $15 per user Team price and free tier. - [Harmonic Security, ranked 3rd of 16](/platforms/harmonic-security/): Harmonic documents hook-based inline blocking of prompts and tool calls across named coding agents (Claude Code, Codex, Cursor, Copilot CLI and others) plus an MCP gateway with identity-aware access groups. ## Sources - github.com/knostic/openclaw-detect: https://github.com/knostic/openclaw-detect (read 12 Nov 2026) - github.com/knostic/openclaw-shield: https://github.com/knostic/openclaw-shield (read 12 Nov 2026) - siliconangle.com/2025/03/05/knostic-raises-11m-strengthen-enterprise-ai-security-need-know-access-controls: https://siliconangle.com/2025/03/05/knostic-raises-11m-strengthen-enterprise-ai-security-need-know-access-controls/ (read 12 Nov 2026) - github.com/knostic/openclaw-telemetry: https://github.com/knostic/openclaw-telemetry (read 12 Nov 2026) - github.com/knostic/AgentSonar: https://github.com/knostic/AgentSonar (read 12 Nov 2026) - github.com/knostic/OpenAnt: https://github.com/knostic/OpenAnt (read 12 Nov 2026) - docs.harmonicsecurity.app/portal-guides/integration-guides/configure-opentelemetry: https://docs.harmonicsecurity.app/portal-guides/integration-guides/configure-opentelemetry (read 12 Nov 2026) - docs.harmonicsecurity.app/portal-guides/integration-guides/configure-anthropic-compliance-api-connector: https://docs.harmonicsecurity.app/portal-guides/integration-guides/configure-anthropic-compliance-api-connector (read 12 Nov 2026) - docs.harmonicsecurity.app/protect-agent/installation-and-deployment/frequently-asked-questions: https://docs.harmonicsecurity.app/protect-agent/installation-and-deployment/frequently-asked-questions (read 12 Nov 2026) - docs.harmonicsecurity.app/portal-guides/detection-and-response/sensitive-data-controls-for-coding-agents-and-desktop-applications: https://docs.harmonicsecurity.app/portal-guides/detection-and-response/sensitive-data-controls-for-coding-agents-and-desktop-applications (read 12 Nov 2026) - docs.harmonicsecurity.app/portal-guides/managing-your-organization/configure-mcp-server-access-controls: https://docs.harmonicsecurity.app/portal-guides/managing-your-organization/configure-mcp-server-access-controls (read 12 Nov 2026) - www.knostic.ai/kirin: https://www.knostic.ai/kirin (read 12 Nov 2026) - www.knostic.ai/faq: https://www.knostic.ai/faq (read 12 Nov 2026) - www.harmonic.security/pricing: https://www.harmonic.security/pricing (read 12 Nov 2026) - docs.harmonicsecurity.app/mcp-gateway/getting-started/what-is-mcp-gateway: https://docs.harmonicsecurity.app/mcp-gateway/getting-started/what-is-mcp-gateway (read 12 Nov 2026) - docs.harmonicsecurity.app/llms.txt: https://docs.harmonicsecurity.app/llms.txt (read 12 Nov 2026) - docs.harmonicsecurity.app/mcp-gateway/cloud-mcp-gateway/creating-a-new-connector.md: https://docs.harmonicsecurity.app/mcp-gateway/cloud-mcp-gateway/creating-a-new-connector.md (read 13 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/compare/knostic-vs-harmonic-security/ --- # Knostic beats Lasso Security 64.0 to 53.0 on the published weights HEAD TO HEAD, read 13 Nov 2026 **The short answer.** Knostic scores 64.0 of 100 against Lasso Security at 53.0 on the published weights. Knostic leads most on Stops (7 to 5), Deployment (7 to 5) and Transparency (7 to 4), while Lasso Security is ahead on Beyond (5 to 3) and Proof (named customers, funding, analyst coverage) (8 to 5). By situation, Knostic leads in all five. ## Verdict on the published weights Knostic scores 64.0 of 100 and Lasso Security scores 53.0. Knostic is ahead on 5 of 7 criteria and behind on 2; 0 are level. Knostic leads on Stops (7 to 5), Deployment (7 to 5), Transparency (7 to 4), Sees (7 to 6) and Policy (6 to 5), while Lasso Security is ahead on Beyond (5 to 3) and Proof (named customers, funding, analyst coverage) (8 to 5). ## Criterion by criterion | Criterion | Weight | Knostic | Lasso Security | | --- | --- | --- | --- | | Sees agentic activity on the endpoint | 25 | 7 | 6 | | Stops the action | 25 | 7 | 5 | | Deployment friction | 15 | 7 | 5 | | Beyond the endpoint | 10 | 3 | 5 | | Policy and governance | 10 | 6 | 5 | | Transparency | 10 | 7 | 4 | | Proof | 5 | 5 | 8 | | Total (0 to 100) | 100 | 64.0 | 53.0 | Knostic against Lasso Security, scores out of 10 on the site weights. ## What each sees and can stop | Tool | Surface strip | Stop strip | | --- | --- | --- | | [Knostic](/platforms/knostic/) | Surfaces seen: Desktop apps documented; IDE assistants documented; Command-line agents documented; MCP servers documented; Browser extensions partial; SaaS copilots partial | Actions it can stop: File read documented; Shell command documented; Network call partial; Credential access documented; Data paste partial; Package install partial | | [Lasso Security](/platforms/lasso-security/) | Surfaces seen: Desktop apps partial; IDE assistants documented; Command-line agents documented; MCP servers documented; Browser extensions none found; SaaS copilots partial | Actions it can stop: File read none found; Shell command partial; Network call partial; Credential access partial; Data paste documented; Package install none found | Filled documented, half partial, open none found. Surface order: desktop apps, IDE assistants, command-line agents, MCP servers, browser extensions, SaaS copilots. Stop order: file read, shell command, network call, credential access, data paste, package install. | Where they differ | Knostic | Lasso Security | | --- | --- | --- | | Desktop apps | documented | partial | | Browser extensions | partial | none found | | Stops: file read | documented | none found | | Stops: shell command | documented | partial | | Stops: credential access | documented | partial | | Stops: data paste | partial | documented | | Stops: package install | partial | none found | 7 of 12 cells differ between Knostic and Lasso Security as documented on 13 Nov 2026. ## Who should pick what, by situation - **Coding agents** ([an engineering organization where every developer runs coding agents with shell access](/best-for/coding-agents/)): Knostic leads Lasso Security 66.0 to 53.5, ahead on Stops (7 to 5), Deployment (7 to 5), Sees (7 to 6), Transparency (7 to 4) and Policy (6 to 5), while Lasso Security is ahead on Proof (named customers, funding, analyst coverage) (8 to 5) and Beyond (5 to 3). - **Regulated data** ([a regulated company that must show which AI tools touch customer data](/best-for/regulated-data/)): Knostic leads Lasso Security 60.5 to 54.0, ahead on Sees (7 to 6), Stops (7 to 5), Policy (6 to 5), Transparency (7 to 4) and Deployment (7 to 5), while Lasso Security is ahead on Beyond (5 to 3) and Proof (named customers, funding, analyst coverage) (8 to 5). - **Lean team with EDR** ([a lean security team with EDR in place and no view of AI agents](/best-for/lean-team-edr/)): Knostic leads Lasso Security 66.0 to 54.0, ahead on Deployment (7 to 5), Stops (7 to 5), Sees (7 to 6), Transparency (7 to 4) and Policy (6 to 5), while Lasso Security is ahead on Proof (named customers, funding, analyst coverage) (8 to 5) and Beyond (5 to 3). - **Staff rollout** ([a company rolling out an AI assistant and browser agents to all staff](/best-for/staff-rollout/)): Knostic leads Lasso Security 58.0 to 53.0, ahead on Stops (7 to 5), Deployment (7 to 5), Sees (7 to 6), Transparency (7 to 4) and Policy (6 to 5), while Lasso Security is ahead on Beyond (5 to 3) and Proof (named customers, funding, analyst coverage) (8 to 5). - **Board in 30 days** ([a CISO asked by the board for an inventory of AI agents within 30 days](/best-for/board-30-days/)): Knostic leads Lasso Security 64.5 to 55.0, ahead on Deployment (7 to 5), Sees (7 to 6), Transparency (7 to 4), Stops (7 to 5) and Policy (6 to 5), while Lasso Security is ahead on Beyond (5 to 3) and Proof (named customers, funding, analyst coverage) (8 to 5). ## What would change this ranking Knostic is 11.0 points clear of Lasso Security on the published weights. The smallest single change that would put Lasso Security first is a Stops score of 10 instead of 5 (weight 25). Scores move only when a sourced fact changes ([how](/method/)). ## Read the profiles - [Knostic, ranked 9th of 16](/platforms/knostic/): Knostic documents Enforce-mode blocking of secret file reads, risky MCP servers, destructive commands and unsafe packages inside agents, with a published $15 per user Team price and free tier. - [Lasso Security, ranked 13th of 16](/platforms/lasso-security/): Lasso documents inspecting every coding-agent tool call before execution via Claude Code, Cursor and Codex hooks, with out-of-scope calls flagged or blocked, plus an open-source MCP gateway. ## Sources - github.com/knostic/openclaw-detect: https://github.com/knostic/openclaw-detect (read 12 Nov 2026) - github.com/knostic/openclaw-shield: https://github.com/knostic/openclaw-shield (read 12 Nov 2026) - siliconangle.com/2025/03/05/knostic-raises-11m-strengthen-enterprise-ai-security-need-know-access-controls: https://siliconangle.com/2025/03/05/knostic-raises-11m-strengthen-enterprise-ai-security-need-know-access-controls/ (read 12 Nov 2026) - github.com/knostic/openclaw-telemetry: https://github.com/knostic/openclaw-telemetry (read 12 Nov 2026) - github.com/knostic/AgentSonar: https://github.com/knostic/AgentSonar (read 12 Nov 2026) - github.com/knostic/OpenAnt: https://github.com/knostic/OpenAnt (read 12 Nov 2026) - github.com/lasso-security/mcp-gateway: https://github.com/lasso-security/mcp-gateway (read 12 Nov 2026) - github.com/lasso-security/kong-ai-lasso-guardrail: https://github.com/lasso-security/kong-ai-lasso-guardrail (read 12 Nov 2026) - github.com/lasso-security/claude-hooks: https://github.com/lasso-security/claude-hooks (read 12 Nov 2026) - www.theregister.com/ai-and-ml/2026/09/17/ai-model-watermarking-changes-agent-behavior/5296998: https://www.theregister.com/ai-and-ml/2026/09/17/ai-model-watermarking-changes-agent-behavior/5296998 (read 12 Nov 2026) - www.knostic.ai/pricing: https://www.knostic.ai/pricing (read 12 Nov 2026) - www.knostic.ai/about: https://www.knostic.ai/about (read 12 Nov 2026) - www.knostic.ai/blog/when-an-mcp-tool-asks-for-your-github-token: https://www.knostic.ai/blog/when-an-mcp-tool-asks-for-your-github-token (read 13 Nov 2026) - docs.knostic.ai (checked, nothing found): https://docs.knostic.ai/ (read 12 Nov 2026) - news.google.com/rss/search?q=%22Knostic%22&hl=en-US&gl=US&ceid=US:en (checked, nothing found): https://news.google.com/rss/search?q=%22Knostic%22&hl=en-US&gl=US&ceid=US:en (read 13 Nov 2026) - www.securityweek.com/knostic-secures-11-million-to-rein-in-enterprise-ai-data-leakage-oversharing: https://www.securityweek.com/knostic-secures-11-million-to-rein-in-enterprise-ai-data-leakage-oversharing/ (read 12 Nov 2026) - finance.yahoo.com/sectors/technology/articles/knostic-named-2026-cb-insights-140000898.html: https://finance.yahoo.com/sectors/technology/articles/knostic-named-2026-cb-insights-140000898.html (read 12 Nov 2026) - www.knostic.ai/blog/gartner-cool-vendor-recognition: https://www.knostic.ai/blog/gartner-cool-vendor-recognition (read 12 Nov 2026) - www.knostic.ai (checked, nothing found): https://www.knostic.ai (read 12 Nov 2026) - docs.lasso.security: https://docs.lasso.security/ (read 12 Nov 2026) - www.lasso.security/pricing (checked, nothing found): https://www.lasso.security/pricing (read 12 Nov 2026) - siliconangle.com/2026/09/02/lasso-security-launches-cpu-only-ai-guardrail-raises-30m: https://siliconangle.com/2026/09/02/lasso-security-launches-cpu-only-ai-guardrail-raises-30m/ (read 12 Nov 2026) - finance.yahoo.com/technology/ai/articles/lasso-security-announces-future-ai-080000059.html: https://finance.yahoo.com/technology/ai/articles/lasso-security-announces-future-ai-080000059.html (read 12 Nov 2026) - www.lasso.security: https://www.lasso.security (read 12 Nov 2026) - techcentral.co.za/neworder-lasso-security-ai-agent-controls/285558: https://techcentral.co.za/neworder-lasso-security-ai-agent-controls/285558/ (read 13 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/compare/knostic-vs-lasso-security/ --- # Zenity beats Pillar Security 65.0 to 53.0 on the published weights HEAD TO HEAD, read 13 Nov 2026 **The short answer.** Zenity scores 65.0 of 100 against Pillar Security at 53.0 on the published weights. Zenity leads most on Beyond (7 to 2), Transparency (7 to 4) and Stops (7 to 6), while Pillar Security is ahead on Deployment (6 to 5). By situation, Zenity leads in all five. ## Verdict on the published weights Zenity scores 65.0 of 100 and Pillar Security scores 53.0. Zenity is ahead on 5 of 7 criteria and behind on 1; 1 is level. Zenity leads on Beyond (7 to 2), Transparency (7 to 4), Stops (7 to 6), Policy (7 to 5) and Proof (named customers, funding, analyst coverage) (8 to 6), while Pillar Security is ahead on Deployment (6 to 5). ## Criterion by criterion | Criterion | Weight | Pillar Security | Zenity | | --- | --- | --- | --- | | Sees agentic activity on the endpoint | 25 | 6 | 6 | | Stops the action | 25 | 6 | 7 | | Deployment friction | 15 | 6 | 5 | | Beyond the endpoint | 10 | 2 | 7 | | Policy and governance | 10 | 5 | 7 | | Transparency | 10 | 4 | 7 | | Proof | 5 | 6 | 8 | | Total (0 to 100) | 100 | 53.0 | 65.0 | Pillar Security against Zenity, scores out of 10 on the site weights. ## What each sees and can stop | Tool | Surface strip | Stop strip | | --- | --- | --- | | [Pillar Security](/platforms/pillar-security/) | Surfaces seen: Desktop apps none found; IDE assistants documented; Command-line agents documented; MCP servers documented; Browser extensions none found; SaaS copilots none found | Actions it can stop: File read documented; Shell command documented; Network call documented; Credential access documented; Data paste none found; Package install none found | | [Zenity](/platforms/zenity/) | Surfaces seen: Desktop apps partial; IDE assistants documented; Command-line agents documented; MCP servers documented; Browser extensions partial; SaaS copilots documented | Actions it can stop: File read partial; Shell command documented; Network call documented; Credential access documented; Data paste partial; Package install partial | Filled documented, half partial, open none found. Surface order: desktop apps, IDE assistants, command-line agents, MCP servers, browser extensions, SaaS copilots. Stop order: file read, shell command, network call, credential access, data paste, package install. | Where they differ | Pillar Security | Zenity | | --- | --- | --- | | Desktop apps | none found | partial | | Browser extensions | none found | partial | | SaaS copilots | none found | documented | | Stops: file read | documented | partial | | Stops: data paste | none found | partial | | Stops: package install | none found | partial | 6 of 12 cells differ between Pillar Security and Zenity as documented on 13 Nov 2026. ## Who should pick what, by situation - **Coding agents** ([an engineering organization where every developer runs coding agents with shell access](/best-for/coding-agents/)): Zenity leads Pillar Security 65.0 to 56.0, ahead on Stops (7 to 6), Beyond (7 to 2), Policy (7 to 5), Transparency (7 to 4) and Proof (named customers, funding, analyst coverage) (8 to 6), while Pillar Security is ahead on Deployment (6 to 5). - **Regulated data** ([a regulated company that must show which AI tools touch customer data](/best-for/regulated-data/)): Zenity leads Pillar Security 66.5 to 50.5, ahead on Beyond (7 to 2), Policy (7 to 5), Stops (7 to 6), Transparency (7 to 4) and Proof (named customers, funding, analyst coverage) (8 to 6), while Pillar Security is ahead on Deployment (6 to 5). - **Lean team with EDR** ([a lean security team with EDR in place and no view of AI agents](/best-for/lean-team-edr/)): Zenity leads Pillar Security 62.5 to 56.0, ahead on Beyond (7 to 2), Stops (7 to 6), Policy (7 to 5), Transparency (7 to 4) and Proof (named customers, funding, analyst coverage) (8 to 6), while Pillar Security is ahead on Deployment (6 to 5). - **Staff rollout** ([a company rolling out an AI assistant and browser agents to all staff](/best-for/staff-rollout/)): Zenity leads Pillar Security 65.5 to 48.0, ahead on Beyond (7 to 2), Stops (7 to 6), Policy (7 to 5), Transparency (7 to 4) and Proof (named customers, funding, analyst coverage) (8 to 6), while Pillar Security is ahead on Deployment (6 to 5). - **Board in 30 days** ([a CISO asked by the board for an inventory of AI agents within 30 days](/best-for/board-30-days/)): Zenity leads Pillar Security 60.5 to 54.5, ahead on Beyond (7 to 2), Transparency (7 to 4), Policy (7 to 5), Proof (named customers, funding, analyst coverage) (8 to 6) and Stops (7 to 6), while Pillar Security is ahead on Deployment (6 to 5). ## What would change this ranking Zenity is 12.0 points clear of Pillar Security on the published weights. No single score change could close that gap. Scores move only when a sourced fact changes ([how](/method/)). ## Read the profiles - [Pillar Security, ranked 14th of 16](/platforms/pillar-security/): Documents pre-action blocking in the agent's own pre-tool hook for Claude Code, Codex CLI and Cursor, with credential-exfiltration and persistence controls and MDM-installed hooks developers cannot disable. - [Zenity, ranked 8th of 16](/platforms/zenity/): Documents pre-execution blocking of coding-agent tool calls through native hooks and an MCP gateway, including credential-read-then-network sequences. ## Sources - thehackernews.com/2026/04/google-fixes-cvss-10-gemini-cli-ci-rce.html: https://thehackernews.com/2026/04/google-fixes-cvss-10-gemini-cli-ci-rce.html (read 12 Nov 2026) - thehackernews.com/2026/10/opencode-flaw-lets-malicious-mcp.html: https://thehackernews.com/2026/10/opencode-flaw-lets-malicious-mcp.html (read 12 Nov 2026) - www.infosecurity-magazine.com/news/two-critical-flaws-in-n8n-ai: https://www.infosecurity-magazine.com/news/two-critical-flaws-in-n8n-ai/ (read 12 Nov 2026) - docs.pillar.security: https://docs.pillar.security (read 12 Nov 2026) - www.calcalistech.com/ctechnews/article/hkjzjkjrfl: https://www.calcalistech.com/ctechnews/article/hkjzjkjrfl (read 12 Nov 2026) - www.theregister.com/security/2026/09/24/salesforce-agentforce-vulns-allowed-0-click-crm-data-theft-anonymous-phishing/5298958: https://www.theregister.com/security/2026/09/24/salesforce-agentforce-vulns-allowed-0-click-crm-data-theft-anonymous-phishing/5298958 (read 12 Nov 2026) - www.securityweek.com/salesbleed-flaws-in-salesforce-agentforce-enabled-zero-click-data-exfiltration: https://www.securityweek.com/salesbleed-flaws-in-salesforce-agentforce-enabled-zero-click-data-exfiltration/ (read 12 Nov 2026) - www.techradar.com/pro/security/experts-warn-malicious-ai-skills-are-hitting-more-victims-than-ever-with-one-family-amassing-1-7-million-downloads: https://www.techradar.com/pro/security/experts-warn-malicious-ai-skills-are-hitting-more-victims-than-ever-with-one-family-amassing-1-7-million-downloads (read 12 Nov 2026) - www.calcalistech.com/ctechnews/article/rsdllsuy3: https://www.calcalistech.com/ctechnews/article/rsdllsuy3 (read 12 Nov 2026) - zenity.io/use-cases/agent-type/coding-personal-agents: https://zenity.io/use-cases/agent-type/coding-personal-agents (read 12 Nov 2026) - www.pillar.security/blog/introducing-pillar-for-ai-coding-agents: https://www.pillar.security/blog/introducing-pillar-for-ai-coding-agents (read 12 Nov 2026) - www.pillar.security/platform/agentic-endpoint (checked, nothing found): https://www.pillar.security/platform/agentic-endpoint (read 12 Nov 2026) - www.pillar.security/blog/inspecting-the-inspector-rce-in-mcps-official-debugging-tool: https://www.pillar.security/blog/inspecting-the-inspector-rce-in-mcps-official-debugging-tool (read 13 Nov 2026) - www.pillar.security/pricing (checked, nothing found): https://www.pillar.security/pricing (read 12 Nov 2026) - zenity.io: https://zenity.io/ (read 12 Nov 2026) - zenity.io/press-release/zenity-raises-125-million-to-secure-the-era-of-1-billion-ai-agents: https://zenity.io/press-release/zenity-raises-125-million-to-secure-the-era-of-1-billion-ai-agents (read 12 Nov 2026) - zenity.io/press-release/zenity-extends-ai-agent-security-and-governance-to-claude-enterprise: https://zenity.io/press-release/zenity-extends-ai-agent-security-and-governance-to-claude-enterprise (read 12 Nov 2026) - zenity.io/blog/zenity-for-claude-tag-securing-the-shared-agent-in-your-slack-workspace: https://zenity.io/blog/zenity-for-claude-tag-securing-the-shared-agent-in-your-slack-workspace (read 13 Nov 2026) - docs.zenity.io (checked, nothing found): https://docs.zenity.io/ (read 13 Nov 2026) - zenity.io/press-release/zenity-labs-discloses-agentcorruption-a-chain-of-aws-agentcore-flaws: https://zenity.io/press-release/zenity-labs-discloses-agentcorruption-a-chain-of-aws-agentcore-flaws (read 13 Nov 2026) - zenity.io/pricing (checked, nothing found): https://zenity.io/pricing (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/compare/pillar-security-vs-zenity/ --- # SentinelOne beats WitnessAI 62.0 to 61.0 on the published weights HEAD TO HEAD, read 13 Nov 2026 **The short answer.** SentinelOne scores 62.0 of 100 against WitnessAI at 61.0 on the published weights. SentinelOne leads on Stops (7 to 6), Beyond (7 to 6) and Transparency (4 to 3), while WitnessAI is ahead on Sees (8 to 7) and Proof (named customers, funding, analyst coverage) (7 to 5). By situation, WitnessAI leads in 3 of 5 and SentinelOne in 2. ## Verdict on the published weights SentinelOne scores 62.0 of 100 and WitnessAI scores 61.0. SentinelOne is ahead on 3 of 7 criteria and behind on 2; 2 are level. SentinelOne leads on Stops (7 to 6), Beyond (7 to 6) and Transparency (4 to 3), while WitnessAI is ahead on Sees (8 to 7) and Proof (named customers, funding, analyst coverage) (7 to 5). ## Criterion by criterion | Criterion | Weight | WitnessAI | SentinelOne | | --- | --- | --- | --- | | Sees agentic activity on the endpoint | 25 | 8 | 7 | | Stops the action | 25 | 6 | 7 | | Deployment friction | 15 | 5 | 5 | | Beyond the endpoint | 10 | 6 | 7 | | Policy and governance | 10 | 6 | 6 | | Transparency | 10 | 3 | 4 | | Proof | 5 | 7 | 5 | | Total (0 to 100) | 100 | 61.0 | 62.0 | WitnessAI against SentinelOne, scores out of 10 on the site weights. ## What each sees and can stop | Tool | Surface strip | Stop strip | | --- | --- | --- | | [WitnessAI](/platforms/witnessai/) | Surfaces seen: Desktop apps documented; IDE assistants documented; Command-line agents partial; MCP servers documented; Browser extensions partial; SaaS copilots partial | Actions it can stop: File read none found; Shell command partial; Network call documented; Credential access partial; Data paste documented; Package install none found | | [SentinelOne](/platforms/sentinelone/) | Surfaces seen: Desktop apps documented; IDE assistants documented; Command-line agents documented; MCP servers documented; Browser extensions documented; SaaS copilots partial | Actions it can stop: File read none found; Shell command partial; Network call partial; Credential access partial; Data paste documented; Package install none found | Filled documented, half partial, open none found. Surface order: desktop apps, IDE assistants, command-line agents, MCP servers, browser extensions, SaaS copilots. Stop order: file read, shell command, network call, credential access, data paste, package install. | Where they differ | WitnessAI | SentinelOne | | --- | --- | --- | | Command-line agents | partial | documented | | Browser extensions | partial | documented | | Stops: network call | documented | partial | 3 of 12 cells differ between WitnessAI and SentinelOne as documented on 13 Nov 2026. ## Who should pick what, by situation - **Coding agents** ([an engineering organization where every developer runs coding agents with shell access](/best-for/coding-agents/)): SentinelOne leads WitnessAI 63.5 to 62.5, ahead on Stops (7 to 6), Beyond (7 to 6) and Transparency (4 to 3), while WitnessAI is ahead on Sees (8 to 7) and Proof (named customers, funding, analyst coverage) (7 to 5). - **Regulated data** ([a regulated company that must show which AI tools touch customer data](/best-for/regulated-data/)): WitnessAI leads SentinelOne 64.5 to 64.0, ahead on Sees (8 to 7) and Proof (named customers, funding, analyst coverage) (7 to 5), while SentinelOne is ahead on Stops (7 to 6), Beyond (7 to 6) and Transparency (4 to 3). - **Lean team with EDR** ([a lean security team with EDR in place and no view of AI agents](/best-for/lean-team-edr/)): WitnessAI leads SentinelOne 62.5 to 61.5, ahead on Sees (8 to 7) and Proof (named customers, funding, analyst coverage) (7 to 5), while SentinelOne is ahead on Stops (7 to 6), Beyond (7 to 6) and Transparency (4 to 3). - **Staff rollout** ([a company rolling out an AI assistant and browser agents to all staff](/best-for/staff-rollout/)): SentinelOne leads WitnessAI 63.5 to 61.5, ahead on Beyond (7 to 6), Stops (7 to 6) and Transparency (4 to 3), while WitnessAI is ahead on Sees (8 to 7) and Proof (named customers, funding, analyst coverage) (7 to 5). - **Board in 30 days** ([a CISO asked by the board for an inventory of AI agents within 30 days](/best-for/board-30-days/)): WitnessAI leads SentinelOne 64.0 to 61.0, ahead on Sees (8 to 7) and Proof (named customers, funding, analyst coverage) (7 to 5), while SentinelOne is ahead on Beyond (7 to 6), Stops (7 to 6) and Transparency (4 to 3). ## What would change this ranking SentinelOne is one point clear of WitnessAI on the published weights. The smallest single change that would put WitnessAI first is a Deployment score of 6 instead of 5 (weight 15). Scores move only when a sourced fact changes ([how](/method/)). ## Read the profiles - [WitnessAI, ranked 11th of 16](/platforms/witnessai/): WitnessAI documents inline blocking of unapproved MCP servers and tool calls across IDEs, chat apps and custom agents, with audit records per blocked call and one policy for humans and agents. - [SentinelOne, ranked 10th of 16](/platforms/sentinelone/): SentinelOne documents an MCP gateway with allow, block, filter or redact actions and runtime tool-call blocking, plus a lightweight agent and extension that discover shadow AI across browsers, IDEs and terminal assistants. ## Sources - siliconangle.com/2026/01/13/witnessai-debuts-agentic-security-enterprises-deploy-autonomous-ai-agents: https://siliconangle.com/2026/01/13/witnessai-debuts-agentic-security-enterprises-deploy-autonomous-ai-agents/ (read 12 Nov 2026) - www.helpnetsecurity.com/2026/06/17/witnessai-agentic-control: https://www.helpnetsecurity.com/2026/06/17/witnessai-agentic-control/ (read 13 Nov 2026) - pulse2.com/witnessai-profile-rick-caccia-interview: https://pulse2.com/witnessai-profile-rick-caccia-interview/ (read 13 Nov 2026) - www.securityweek.com/witnessai-raises-58-million-for-ai-security-platform: https://www.securityweek.com/witnessai-raises-58-million-for-ai-security-platform/ (read 12 Nov 2026) - siliconangle.com/2026/03/23/sentinelone-snyk-introduce-new-tools-securing-ai-agents: https://siliconangle.com/2026/03/23/sentinelone-snyk-introduce-new-tools-securing-ai-agents/ (read 12 Nov 2026) - chromewebstore.google.com/search/prompt%20security: https://chromewebstore.google.com/search/prompt%20security (read 12 Nov 2026) - www.crn.com/news/security/2026/sentinelone-s-chris-corde-on-why-traditional-edr-is-flying-blind-on-ai-agents: https://www.crn.com/news/security/2026/sentinelone-s-chris-corde-on-why-traditional-edr-is-flying-blind-on-ai-agents (read 13 Nov 2026) - github.com/prompt-security/clawsec: https://github.com/prompt-security/clawsec (read 12 Nov 2026) - github.com/prompt-security: https://github.com/prompt-security (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/compare/witnessai-vs-sentinelone/ --- # Aim Security beats Palo Alto Networks 69.0 to 45.0 on the published weights HEAD TO HEAD, read 13 Nov 2026 **The short answer.** Aim Security scores 69.0 of 100 against Palo Alto Networks at 45.0 on the published weights. Aim Security leads most on Sees (8 to 3), Stops (7 to 5) and Transparency (7 to 4). By situation, Aim Security leads in all five. Aim Security is ahead on 6 of 7 criteria. ## Verdict on the published weights Aim Security scores 69.0 of 100 and Palo Alto Networks scores 45.0. Aim Security is ahead on 6 of 7 criteria and behind on 0; 1 is level. Aim Security leads on Sees (8 to 3), Stops (7 to 5), Transparency (7 to 4), Deployment (5 to 4), Beyond (7 to 6) and Policy (6 to 5). ## Criterion by criterion | Criterion | Weight | Aim Security | Palo Alto Networks | | --- | --- | --- | --- | | Sees agentic activity on the endpoint | 25 | 8 | 3 | | Stops the action | 25 | 7 | 5 | | Deployment friction | 15 | 5 | 4 | | Beyond the endpoint | 10 | 7 | 6 | | Policy and governance | 10 | 6 | 5 | | Transparency | 10 | 7 | 4 | | Proof | 5 | 8 | 8 | | Total (0 to 100) | 100 | 69.0 | 45.0 | Aim Security against Palo Alto Networks, scores out of 10 on the site weights. ## What each sees and can stop | Tool | Surface strip | Stop strip | | --- | --- | --- | | [Aim Security](/platforms/aim-security/) | Surfaces seen: Desktop apps documented; IDE assistants partial; Command-line agents documented; MCP servers documented; Browser extensions documented; SaaS copilots partial | Actions it can stop: File read partial; Shell command partial; Network call partial; Credential access none found; Data paste documented; Package install none found | | [Palo Alto Networks](/platforms/palo-alto-networks/) | Surfaces seen: Desktop apps partial; IDE assistants documented; Command-line agents partial; MCP servers none found; Browser extensions documented; SaaS copilots partial | Actions it can stop: File read none found; Shell command partial; Network call partial; Credential access none found; Data paste documented; Package install none found | Filled documented, half partial, open none found. Surface order: desktop apps, IDE assistants, command-line agents, MCP servers, browser extensions, SaaS copilots. Stop order: file read, shell command, network call, credential access, data paste, package install. | Where they differ | Aim Security | Palo Alto Networks | | --- | --- | --- | | Desktop apps | documented | partial | | IDE assistants | partial | documented | | Command-line agents | documented | partial | | MCP servers | documented | none found | | Stops: file read | partial | none found | 5 of 12 cells differ between Aim Security and Palo Alto Networks as documented on 13 Nov 2026. ## Who should pick what, by situation - **Coding agents** ([an engineering organization where every developer runs coding agents with shell access](/best-for/coding-agents/)): Aim Security leads Palo Alto Networks 69.0 to 45.0, ahead on Sees (8 to 3), Stops (7 to 5), Deployment (5 to 4), Transparency (7 to 4), Policy (6 to 5) and Beyond (7 to 6). - **Regulated data** ([a regulated company that must show which AI tools touch customer data](/best-for/regulated-data/)): Aim Security leads Palo Alto Networks 70.0 to 46.0, ahead on Sees (8 to 3), Stops (7 to 5), Policy (6 to 5), Beyond (7 to 6), Transparency (7 to 4) and Deployment (5 to 4). - **Lean team with EDR** ([a lean security team with EDR in place and no view of AI agents](/best-for/lean-team-edr/)): Aim Security leads Palo Alto Networks 67.5 to 43.0, ahead on Sees (8 to 3), Stops (7 to 5), Deployment (5 to 4), Transparency (7 to 4), Policy (6 to 5) and Beyond (7 to 6). - **Staff rollout** ([a company rolling out an AI assistant and browser agents to all staff](/best-for/staff-rollout/)): Aim Security leads Palo Alto Networks 68.5 to 48.0, ahead on Sees (8 to 3), Stops (7 to 5), Beyond (7 to 6), Deployment (5 to 4), Transparency (7 to 4) and Policy (6 to 5). - **Board in 30 days** ([a CISO asked by the board for an inventory of AI agents within 30 days](/best-for/board-30-days/)): Aim Security leads Palo Alto Networks 68.0 to 41.0, ahead on Sees (8 to 3), Deployment (5 to 4), Transparency (7 to 4), Stops (7 to 5), Beyond (7 to 6) and Policy (6 to 5). ## What would change this ranking Aim Security is 24.0 points clear of Palo Alto Networks on the published weights. No single score change could close that gap. Scores move only when a sourced fact changes ([how](/method/)). ## Read the profiles - [Aim Security, ranked 4th of 16](/platforms/aim-security/): Cato AI Endpoint documents a lightweight client that finds local agents with their MCP servers and skills and blocks an unapproved tool call before it runs, in one dashboard with inline network inspection. - [Palo Alto Networks, ranked 16th of 16](/platforms/palo-alto-networks/): Palo Alto Networks documents both a browser-level control point (extension inventory, copy-paste and upload limits for unsanctioned AI apps) and endpoint coding-agent policy enforcement under Prisma AIRS, backed by public docs and analyst recognition. ## Sources - knowledge.catonetworks.com/docs/product-updates-october-19-2026.md: https://knowledge.catonetworks.com/docs/product-updates-october-19-2026.md (read 13 Nov 2026) - knowledge.catonetworks.com/docs/product-updates-november-9-2026.md: https://knowledge.catonetworks.com/docs/product-updates-november-9-2026.md (read 13 Nov 2026) - knowledge.catonetworks.com/docs/integrating-anthropic-compliance-api-with-cato-ai-security: https://knowledge.catonetworks.com/docs/integrating-anthropic-compliance-api-with-cato-ai-security (read 12 Nov 2026) - techcrunch.com/2024/01/31/aim-security-raises-10m-for-its-genai-security-platform: https://techcrunch.com/2024/01/31/aim-security-raises-10m-for-its-genai-security-platform/ (read 12 Nov 2026) - candid.technology/best-shadow-agent-monitoring-solutions-for-enterprises: https://candid.technology/best-shadow-agent-monitoring-solutions-for-enterprises/ (read 13 Nov 2026) - docs.paloaltonetworks.com/ai-runtime-security/administration/agent-discovery/ai-agent-discovery-limitations: https://docs.paloaltonetworks.com/ai-runtime-security/administration/agent-discovery/ai-agent-discovery-limitations (read 13 Nov 2026) - www.helpnetsecurity.com/2026/09/29/palo-alto-networks-nvidia-ai-agent-security: https://www.helpnetsecurity.com/2026/09/29/palo-alto-networks-nvidia-ai-agent-security/ (read 13 Nov 2026) - siliconangle.com/2026/08/20/zero-networks-expands-palo-alto-networks-integration-to-ai-agent-control: https://siliconangle.com/2026/08/20/zero-networks-expands-palo-alto-networks-integration-to-ai-agent-control/ (read 13 Nov 2026) - www.catonetworks.com/platform/cato-ai-endpoint: https://www.catonetworks.com/platform/cato-ai-endpoint/ (read 12 Nov 2026) - www.prnewswire.com/news-releases/cato-networks-launches-cato-ai-endpoint-to-secure-ai-use-beyond-the-browser-302919573.html: https://www.prnewswire.com/news-releases/cato-networks-launches-cato-ai-endpoint-to-secure-ai-use-beyond-the-browser-302919573.html (read 12 Nov 2026) - knowledge.catonetworks.com/docs/2026-release-notes: https://knowledge.catonetworks.com/docs/2026-release-notes (read 12 Nov 2026) - services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2025-32711: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2025-32711 (read 12 Nov 2026) - thehackernews.com/2025/06/zero-click-ai-vulnerability-exposes.html: https://thehackernews.com/2025/06/zero-click-ai-vulnerability-exposes.html (read 12 Nov 2026) - www.securityweek.com/echoleak-ai-attack-enabled-theft-of-sensitive-data-via-microsoft-365-copilot: https://www.securityweek.com/echoleak-ai-attack-enabled-theft-of-sensitive-data-via-microsoft-365-copilot/ (read 12 Nov 2026) - www.paloaltonetworks.com/prisma/prisma-ai-runtime-security: https://www.paloaltonetworks.com/prisma/prisma-ai-runtime-security (read 12 Nov 2026) - www.paloaltonetworks.com/sase/prisma-browser: https://www.paloaltonetworks.com/sase/prisma-browser (read 12 Nov 2026) - news.google.com/rss/search?q=%22Palo+Alto+Networks%22+(inventory+OR+discovery+OR+MCP+OR+%22coding+agents%22) (checked, nothing found): https://news.google.com/rss/search?q=%22Palo+Alto+Networks%22+(inventory+OR+discovery+OR+MCP+OR+%22coding+agents%22) (read 13 Nov 2026) - docs.paloaltonetworks.com/ai-runtime-security: https://docs.paloaltonetworks.com/ai-runtime-security (read 12 Nov 2026) - docs.paloaltonetworks.com/ai-runtime-security/release-notes: https://docs.paloaltonetworks.com/ai-runtime-security/release-notes (read 12 Nov 2026) - docs.paloaltonetworks.com/prisma-access-browser: https://docs.paloaltonetworks.com/prisma-access-browser (read 13 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/compare/aim-security-vs-palo-alto-networks/ --- # 15 head-to-heads, each decided by the published weights COMPARE, read 13 Nov 2026 **The short answer.** 15 head-to-head comparisons of AI endpoint security tools are published here. Each shows the verdict on the published weights, a criterion table, the surface and stop strips side by side, and a verdict for each of the five buyer situations. Scores run 0 to 100 on weights fixed before any vendor was scored. | Pair | Leader | Totals | | --- | --- | --- | | [Bay vs Island](/compare/bay-vs-island/) | Bay | 71.5 to 66.0 | | [Bay vs Microsoft](/compare/bay-vs-microsoft/) | Microsoft | 75.0 to 71.5 | | [Bay vs Knostic](/compare/bay-vs-knostic/) | Bay | 71.5 to 64.0 | | [Bay vs WitnessAI](/compare/bay-vs-witnessai/) | Bay | 71.5 to 61.0 | | [Bay vs Pillar Security](/compare/bay-vs-pillar-security/) | Bay | 71.5 to 53.0 | | [Bay vs Aim Security](/compare/bay-vs-aim-security/) | Bay | 71.5 to 69.0 | | [Island vs LayerX](/compare/island-vs-layerx/) | Island | 66.0 to 50.0 | | [Island vs Palo Alto Networks](/compare/island-vs-palo-alto-networks/) | Island | 66.0 to 45.0 | | [Microsoft vs CrowdStrike](/compare/microsoft-vs-crowdstrike/) | Microsoft | 75.0 to 66.0 | | [Microsoft vs SentinelOne](/compare/microsoft-vs-sentinelone/) | Microsoft | 75.0 to 62.0 | | [Knostic vs Harmonic Security](/compare/knostic-vs-harmonic-security/) | Harmonic Security | 71.5 to 64.0 | | [Knostic vs Lasso Security](/compare/knostic-vs-lasso-security/) | Knostic | 64.0 to 53.0 | | [Pillar Security vs Zenity](/compare/pillar-security-vs-zenity/) | Zenity | 65.0 to 53.0 | | [WitnessAI vs SentinelOne](/compare/witnessai-vs-sentinelone/) | SentinelOne | 62.0 to 61.0 | | [Aim Security vs Palo Alto Networks](/compare/aim-security-vs-palo-alto-networks/) | Aim Security | 69.0 to 45.0 | Leader on the site weights. **About Bay.** Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). ## Sources - Scores dataset, CSV: https://aiendpointsecurityranked.com/datasets/aes-scores.csv (read 13 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/compare/ --- # Bay and Microsoft tie for engineering teams running coding agents at 75.5 BEST FOR, read 13 Nov 2026 **The short answer.** For an engineering organization where every developer runs coding agents with shell access, Bay and Microsoft tie for first of 16 at 75.5 and both score 8 on Stops, the heaviest criterion here, so name order puts Bay 1st. Both are ahead of Harmonic Security at 72.0. ## The ranking for this situation | Rank | Tool | Total | Sees (25) | Stops (35) | Deployment (15) | Beyond (5) | Policy (10) | Transparency (5) | Proof (5) | | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | | 1 | [Bay](/platforms/bay/) | 75.5 | 8 | 8 | 9 | 7 | 6 | 1 | 8 | | 2 | [Microsoft](/platforms/microsoft/) | 75.5 | 8 | 8 | 6 | 8 | 9 | 7 | 4 | | 3 | [Harmonic Security](/platforms/harmonic-security/) | 72.0 | 8 | 7 | 7 | 7 | 7 | 6 | 7 | | 4 | [Noma Security](/platforms/noma-security/) | 69.5 | 8 | 8 | 6 | 5 | 5 | 4 | 6 | | 5 | [Aim Security](/platforms/aim-security/) | 69.0 | 8 | 7 | 5 | 7 | 6 | 7 | 8 | | 6 | [CrowdStrike](/platforms/crowdstrike/) | 68.5 | 8 | 7 | 6 | 6 | 7 | 3 | 7 | | 7 | [Island](/platforms/island/) | 67.5 | 7 | 7 | 6 | 8 | 7 | 3 | 8 | | 8 | [Knostic](/platforms/knostic/) | 66.0 | 7 | 7 | 7 | 3 | 6 | 7 | 5 | | 9 | [Zenity](/platforms/zenity/) | 65.0 | 6 | 7 | 5 | 7 | 7 | 7 | 8 | | 10 | [SentinelOne](/platforms/sentinelone/) | 63.5 | 7 | 7 | 5 | 7 | 6 | 4 | 5 | | 11 | [WitnessAI](/platforms/witnessai/) | 62.5 | 8 | 6 | 5 | 6 | 6 | 3 | 7 | | 12 | [Pillar Security](/platforms/pillar-security/) | 56.0 | 6 | 6 | 6 | 2 | 5 | 4 | 6 | | 13 | [Koi](/platforms/koi/) | 55.5 | 7 | 5 | 6 | 2 | 4 | 6 | 7 | | 14 | [Lasso Security](/platforms/lasso-security/) | 53.5 | 6 | 5 | 5 | 5 | 5 | 4 | 8 | | 15 | [LayerX](/platforms/layerx/) | 50.5 | 5 | 5 | 5 | 5 | 6 | 4 | 5 | | 16 | [Palo Alto Networks](/platforms/palo-alto-networks/) | 45.0 | 3 | 5 | 4 | 6 | 5 | 4 | 8 | All 16 tools on this situation's weights. | Tied at | Tools in order | Stops (heaviest) | What sets the order | | --- | --- | --- | --- | | 75.5 | Bay, Microsoft | 8, 8 | name order, as the scores are equal | Tied totals break on the higher score on the heaviest criterion, then by name. ## The weights for this situation A coding agent with a shell can delete, exfiltrate or install in seconds, so stopping the action before it completes matters most, followed by seeing which agents, MCP servers and extensions each developer runs. | Criterion | This situation | Site index | | --- | --- | --- | | Sees agentic activity on the endpoint | 25 | 25 | | Stops the action | 35 | 25 | | Deployment friction | 15 | 15 | | Beyond the endpoint | 5 | 10 | | Policy and governance | 10 | 10 | | Transparency | 5 | 10 | | Proof | 5 | 5 | Fixed 12 Nov 2026 before any score. ## Verdict Bay and Microsoft tie at 75.5 and both score 8 on Stops, the heaviest criterion here, so name order puts Bay 1st. Both are ahead of Harmonic Security at 72.0. Bay leads on Deployment (9 to 6) and Proof (named customers, funding, analyst coverage) (8 to 4), while Microsoft is ahead on Policy (9 to 6), Transparency (7 to 1) and Beyond (8 to 7). Bay ranks first here at 75.5, level with Microsoft at 75.5. Bay is a young company that left stealth in 2026, publishes no price and has little independent testing behind its claims. ## Strength of the top three - [Bay](/platforms/bay/), 75.5: Bay documents on-device, session-aware allow/ask/deny/steer enforcement with named action types, agentless MDM/EDR deployment in under 5 minutes (Bay's claim), and one policy across endpoint, browser and cloud. - [Microsoft](/platforms/microsoft/), 75.5: Defender discovers dozens of local agents and MCP servers on Windows and macOS (GA) and can block prompt-injection-driven actions in Claude Code, Codex CLI and GitHub Copilot CLI on Windows, with Endpoint DLP blocking pastes and uploads to AI sites. - [Harmonic Security](/platforms/harmonic-security/), 72.0: Harmonic documents hook-based inline blocking of prompts and tool calls across named coding agents (Claude Code, Codex, Cursor, Copilot CLI and others) plus an MCP gateway with identity-aware access groups. ## Who should pick what - Pick: Bay at 75.5, first on the published tie rule. - Runner-up: Microsoft at 75.5. Other buyer situations: [all five, with their leaders](/best-for/). ## What would change this ranking Bay and Microsoft share the top total for engineering teams running coding agents, so any one-point gain for Microsoft on a weighted criterion would reverse the order. Scores move only when a sourced fact changes ([how](/method/)). ## Sources - softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control: https://softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control (read 12 Nov 2026) - siliconangle.com/2026/11/10/ai-security-startup-bay-launches-with-50m-to-vet-agent-actions-before-they-run: https://siliconangle.com/2026/11/10/ai-security-startup-bay-launches-with-50m-to-vet-agent-actions-before-they-run/ (read 12 Nov 2026) - learn.microsoft.com/en-us/defender-endpoint/local-agent-discovery-overview: https://learn.microsoft.com/en-us/defender-endpoint/local-agent-discovery-overview (read 12 Nov 2026) - learn.microsoft.com/en-us/defender-endpoint/ai-agent-support-matrix: https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-support-matrix (read 12 Nov 2026) - learn.microsoft.com/en-us/defender-endpoint/whats-new-in-microsoft-defender-endpoint: https://learn.microsoft.com/en-us/defender-endpoint/whats-new-in-microsoft-defender-endpoint (read 12 Nov 2026) - docs.harmonicsecurity.app/portal-guides/integration-guides/configure-opentelemetry: https://docs.harmonicsecurity.app/portal-guides/integration-guides/configure-opentelemetry (read 12 Nov 2026) - docs.harmonicsecurity.app/portal-guides/integration-guides/configure-anthropic-compliance-api-connector: https://docs.harmonicsecurity.app/portal-guides/integration-guides/configure-anthropic-compliance-api-connector (read 12 Nov 2026) - docs.harmonicsecurity.app/protect-agent/installation-and-deployment/frequently-asked-questions: https://docs.harmonicsecurity.app/protect-agent/installation-and-deployment/frequently-asked-questions (read 12 Nov 2026) - docs.harmonicsecurity.app/portal-guides/detection-and-response/sensitive-data-controls-for-coding-agents-and-desktop-applications: https://docs.harmonicsecurity.app/portal-guides/detection-and-response/sensitive-data-controls-for-coding-agents-and-desktop-applications (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/best-for/coding-agents/ --- # Microsoft leads for regulated companies, 79.0 to Harmonic Security's 72.5 BEST FOR, read 13 Nov 2026 **The short answer.** For a regulated company that must show which AI tools touch customer data, Microsoft ranks first of 16 at 79.0 of 100, ahead of Harmonic Security at 72.5 and Bay at 70.5. Scores run 0 to 100 on weights fixed before any vendor was scored. ## The ranking for this situation | Rank | Tool | Total | Sees (30) | Stops (15) | Deployment (5) | Beyond (15) | Policy (25) | Transparency (5) | Proof (5) | | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | | 1 | [Microsoft](/platforms/microsoft/) | 79.0 | 8 | 8 | 6 | 8 | 9 | 7 | 4 | | 2 | [Harmonic Security](/platforms/harmonic-security/) | 72.5 | 8 | 7 | 7 | 7 | 7 | 6 | 7 | | 3 | [Bay](/platforms/bay/) | 70.5 | 8 | 8 | 9 | 7 | 6 | 1 | 8 | | 4 | [Aim Security](/platforms/aim-security/) | 70.0 | 8 | 7 | 5 | 7 | 6 | 7 | 8 | | 5 | [Island](/platforms/island/) | 69.5 | 7 | 7 | 6 | 8 | 7 | 3 | 8 | | 6 | [CrowdStrike](/platforms/crowdstrike/) | 69.0 | 8 | 7 | 6 | 6 | 7 | 3 | 7 | | 7 | [Zenity](/platforms/zenity/) | 66.5 | 6 | 7 | 5 | 7 | 7 | 7 | 8 | | 8 | [WitnessAI](/platforms/witnessai/) | 64.5 | 8 | 6 | 5 | 6 | 6 | 3 | 7 | | 9 | [Noma Security](/platforms/noma-security/) | 64.0 | 8 | 8 | 6 | 5 | 5 | 4 | 6 | | 10 | [SentinelOne](/platforms/sentinelone/) | 64.0 | 7 | 7 | 5 | 7 | 6 | 4 | 5 | | 11 | [Knostic](/platforms/knostic/) | 60.5 | 7 | 7 | 7 | 3 | 6 | 7 | 5 | | 12 | [Lasso Security](/platforms/lasso-security/) | 54.0 | 6 | 5 | 5 | 5 | 5 | 4 | 8 | | 13 | [LayerX](/platforms/layerx/) | 52.0 | 5 | 5 | 5 | 5 | 6 | 4 | 5 | | 14 | [Koi](/platforms/koi/) | 51.0 | 7 | 5 | 6 | 2 | 4 | 6 | 7 | | 15 | [Pillar Security](/platforms/pillar-security/) | 50.5 | 6 | 6 | 6 | 2 | 5 | 4 | 6 | | 16 | [Palo Alto Networks](/platforms/palo-alto-networks/) | 46.0 | 3 | 5 | 4 | 6 | 5 | 4 | 8 | All 16 tools on this situation's weights. | Tied at | Tools in order | Sees (heaviest) | What sets the order | | --- | --- | --- | --- | | 64.0 | Noma Security, SentinelOne | 8, 7 | the higher Sees score | Tied totals break on the higher score on the heaviest criterion, then by name. ## The weights for this situation An auditor asks for the inventory and the evidence trail, so seeing every AI tool and keeping a policy record with an audit log outweigh how fast the tool installs. | Criterion | This situation | Site index | | --- | --- | --- | | Sees agentic activity on the endpoint | 30 | 25 | | Stops the action | 15 | 25 | | Deployment friction | 5 | 15 | | Beyond the endpoint | 15 | 10 | | Policy and governance | 25 | 10 | | Transparency | 5 | 10 | | Proof | 5 | 5 | Fixed 12 Nov 2026 before any score. ## Verdict Microsoft leads at 79.0 of 100, ahead of Harmonic Security at 72.5 and Bay at 70.5. Microsoft leads on Policy (9 to 7), Stops (8 to 7), Beyond (8 to 7) and Transparency (7 to 6), while Harmonic Security is ahead on Proof (named customers, funding, analyst coverage) (7 to 4) and Deployment (7 to 6). Bay ranks 3rd of 16 at 70.5, behind Microsoft at 79.0. Microsoft's lead over Bay comes from policy and governance (9 to 6) and transparency (7 to 1). ## Strength of the top three - [Microsoft](/platforms/microsoft/), 79.0: Defender discovers dozens of local agents and MCP servers on Windows and macOS (GA) and can block prompt-injection-driven actions in Claude Code, Codex CLI and GitHub Copilot CLI on Windows, with Endpoint DLP blocking pastes and uploads to AI sites. - [Harmonic Security](/platforms/harmonic-security/), 72.5: Harmonic documents hook-based inline blocking of prompts and tool calls across named coding agents (Claude Code, Codex, Cursor, Copilot CLI and others) plus an MCP gateway with identity-aware access groups. - [Bay](/platforms/bay/), 70.5: Bay documents on-device, session-aware allow/ask/deny/steer enforcement with named action types, agentless MDM/EDR deployment in under 5 minutes (Bay's claim), and one policy across endpoint, browser and cloud. ## Who should pick what - Pick: Microsoft at 79.0. - Runner-up: Harmonic Security at 72.5. Other buyer situations: [all five, with their leaders](/best-for/). ## What would change this ranking Microsoft is 6.5 points clear of Harmonic Security for regulated companies. The smallest single change that would put Harmonic Security first is a Policy score of 10 instead of 7 (weight 25). Bay would move up one place, past Harmonic Security, with a Policy score of 7 instead of 6. Scores move only when a sourced fact changes ([how](/method/)). ## Sources - learn.microsoft.com/en-us/defender-endpoint/local-agent-discovery-overview: https://learn.microsoft.com/en-us/defender-endpoint/local-agent-discovery-overview (read 12 Nov 2026) - learn.microsoft.com/en-us/defender-endpoint/ai-agent-support-matrix: https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-support-matrix (read 12 Nov 2026) - learn.microsoft.com/en-us/defender-endpoint/whats-new-in-microsoft-defender-endpoint: https://learn.microsoft.com/en-us/defender-endpoint/whats-new-in-microsoft-defender-endpoint (read 12 Nov 2026) - docs.harmonicsecurity.app/portal-guides/integration-guides/configure-opentelemetry: https://docs.harmonicsecurity.app/portal-guides/integration-guides/configure-opentelemetry (read 12 Nov 2026) - docs.harmonicsecurity.app/portal-guides/integration-guides/configure-anthropic-compliance-api-connector: https://docs.harmonicsecurity.app/portal-guides/integration-guides/configure-anthropic-compliance-api-connector (read 12 Nov 2026) - docs.harmonicsecurity.app/protect-agent/installation-and-deployment/frequently-asked-questions: https://docs.harmonicsecurity.app/protect-agent/installation-and-deployment/frequently-asked-questions (read 12 Nov 2026) - docs.harmonicsecurity.app/portal-guides/detection-and-response/sensitive-data-controls-for-coding-agents-and-desktop-applications: https://docs.harmonicsecurity.app/portal-guides/detection-and-response/sensitive-data-controls-for-coding-agents-and-desktop-applications (read 12 Nov 2026) - softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control: https://softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control (read 12 Nov 2026) - siliconangle.com/2026/11/10/ai-security-startup-bay-launches-with-50m-to-vet-agent-actions-before-they-run: https://siliconangle.com/2026/11/10/ai-security-startup-bay-launches-with-50m-to-vet-agent-actions-before-they-run/ (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/best-for/regulated-data/ --- # Bay leads for lean teams that already run EDR, 76.5 to Microsoft's 73.5 BEST FOR, read 13 Nov 2026 **The short answer.** For a lean security team with EDR in place and no view of AI agents, Bay ranks first of 16 at 76.5 of 100, ahead of Microsoft at 73.5 and Harmonic Security at 72.5. Scores run 0 to 100 on weights fixed before any vendor was scored. ## The ranking for this situation | Rank | Tool | Total | Sees (30) | Stops (20) | Deployment (25) | Beyond (5) | Policy (10) | Transparency (5) | Proof (5) | | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | | 1 | [Bay](/platforms/bay/) | 76.5 | 8 | 8 | 9 | 7 | 6 | 1 | 8 | | 2 | [Microsoft](/platforms/microsoft/) | 73.5 | 8 | 8 | 6 | 8 | 9 | 7 | 4 | | 3 | [Harmonic Security](/platforms/harmonic-security/) | 72.5 | 8 | 7 | 7 | 7 | 7 | 6 | 7 | | 4 | [CrowdStrike](/platforms/crowdstrike/) | 68.0 | 8 | 7 | 6 | 6 | 7 | 3 | 7 | | 5 | [Aim Security](/platforms/aim-security/) | 67.5 | 8 | 7 | 5 | 7 | 6 | 7 | 8 | | 6 | [Noma Security](/platforms/noma-security/) | 67.5 | 8 | 8 | 6 | 5 | 5 | 4 | 6 | | 7 | [Island](/platforms/island/) | 66.5 | 7 | 7 | 6 | 8 | 7 | 3 | 8 | | 8 | [Knostic](/platforms/knostic/) | 66.0 | 7 | 7 | 7 | 3 | 6 | 7 | 5 | | 9 | [WitnessAI](/platforms/witnessai/) | 62.5 | 8 | 6 | 5 | 6 | 6 | 3 | 7 | | 10 | [Zenity](/platforms/zenity/) | 62.5 | 6 | 7 | 5 | 7 | 7 | 7 | 8 | | 11 | [SentinelOne](/platforms/sentinelone/) | 61.5 | 7 | 7 | 5 | 7 | 6 | 4 | 5 | | 12 | [Koi](/platforms/koi/) | 57.5 | 7 | 5 | 6 | 2 | 4 | 6 | 7 | | 13 | [Pillar Security](/platforms/pillar-security/) | 56.0 | 6 | 6 | 6 | 2 | 5 | 4 | 6 | | 14 | [Lasso Security](/platforms/lasso-security/) | 54.0 | 6 | 5 | 5 | 5 | 5 | 4 | 8 | | 15 | [LayerX](/platforms/layerx/) | 50.5 | 5 | 5 | 5 | 5 | 6 | 4 | 5 | | 16 | [Palo Alto Networks](/platforms/palo-alto-networks/) | 43.0 | 3 | 5 | 4 | 6 | 5 | 4 | 8 | All 16 tools on this situation's weights. | Tied at | Tools in order | Sees (heaviest) | What sets the order | | --- | --- | --- | --- | | 67.5 | Aim Security, Noma Security | 8, 8 | name order, as the scores are equal | | 62.5 | WitnessAI, Zenity | 8, 6 | the higher Sees score | Tied totals break on the higher score on the heaviest criterion, then by name. ## The weights for this situation A small team needs the missing view quickly without another heavy agent to run, so visibility and low deployment friction carry the most weight. | Criterion | This situation | Site index | | --- | --- | --- | | Sees agentic activity on the endpoint | 30 | 25 | | Stops the action | 20 | 25 | | Deployment friction | 25 | 15 | | Beyond the endpoint | 5 | 10 | | Policy and governance | 10 | 10 | | Transparency | 5 | 10 | | Proof | 5 | 5 | Fixed 12 Nov 2026 before any score. ## Verdict Bay leads at 76.5 of 100, ahead of Microsoft at 73.5 and Harmonic Security at 72.5. Bay leads on Deployment (9 to 6) and Proof (named customers, funding, analyst coverage) (8 to 4), while Microsoft is ahead on Policy (9 to 6), Transparency (7 to 1) and Beyond (8 to 7). Bay ranks first here at 76.5, ahead of Microsoft at 73.5. Bay is a young company that left stealth in 2026, publishes no price and has little independent testing behind its claims. ## Strength of the top three - [Bay](/platforms/bay/), 76.5: Bay documents on-device, session-aware allow/ask/deny/steer enforcement with named action types, agentless MDM/EDR deployment in under 5 minutes (Bay's claim), and one policy across endpoint, browser and cloud. - [Microsoft](/platforms/microsoft/), 73.5: Defender discovers dozens of local agents and MCP servers on Windows and macOS (GA) and can block prompt-injection-driven actions in Claude Code, Codex CLI and GitHub Copilot CLI on Windows, with Endpoint DLP blocking pastes and uploads to AI sites. - [Harmonic Security](/platforms/harmonic-security/), 72.5: Harmonic documents hook-based inline blocking of prompts and tool calls across named coding agents (Claude Code, Codex, Cursor, Copilot CLI and others) plus an MCP gateway with identity-aware access groups. ## Who should pick what - Pick: Bay at 76.5. - Runner-up: Microsoft at 73.5. Other buyer situations: [all five, with their leaders](/best-for/). ## What would change this ranking Bay is 3.0 points clear of Microsoft for lean teams that already run EDR. The smallest single change that would put Microsoft first is a Stops score of 10 instead of 8 (weight 20). Scores move only when a sourced fact changes ([how](/method/)). ## Sources - softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control: https://softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control (read 12 Nov 2026) - siliconangle.com/2026/11/10/ai-security-startup-bay-launches-with-50m-to-vet-agent-actions-before-they-run: https://siliconangle.com/2026/11/10/ai-security-startup-bay-launches-with-50m-to-vet-agent-actions-before-they-run/ (read 12 Nov 2026) - learn.microsoft.com/en-us/defender-endpoint/local-agent-discovery-overview: https://learn.microsoft.com/en-us/defender-endpoint/local-agent-discovery-overview (read 12 Nov 2026) - learn.microsoft.com/en-us/defender-endpoint/ai-agent-support-matrix: https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-support-matrix (read 12 Nov 2026) - learn.microsoft.com/en-us/defender-endpoint/whats-new-in-microsoft-defender-endpoint: https://learn.microsoft.com/en-us/defender-endpoint/whats-new-in-microsoft-defender-endpoint (read 12 Nov 2026) - docs.harmonicsecurity.app/portal-guides/integration-guides/configure-opentelemetry: https://docs.harmonicsecurity.app/portal-guides/integration-guides/configure-opentelemetry (read 12 Nov 2026) - docs.harmonicsecurity.app/portal-guides/integration-guides/configure-anthropic-compliance-api-connector: https://docs.harmonicsecurity.app/portal-guides/integration-guides/configure-anthropic-compliance-api-connector (read 12 Nov 2026) - docs.harmonicsecurity.app/protect-agent/installation-and-deployment/frequently-asked-questions: https://docs.harmonicsecurity.app/protect-agent/installation-and-deployment/frequently-asked-questions (read 12 Nov 2026) - docs.harmonicsecurity.app/portal-guides/detection-and-response/sensitive-data-controls-for-coding-agents-and-desktop-applications: https://docs.harmonicsecurity.app/portal-guides/detection-and-response/sensitive-data-controls-for-coding-agents-and-desktop-applications (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/best-for/lean-team-edr/ --- # Microsoft leads for company-wide AI rollouts, 75.5 to Bay's 73.5 BEST FOR, read 13 Nov 2026 **The short answer.** For a company rolling out an AI assistant and browser agents to all staff, Microsoft ranks first of 16 at 75.5 of 100, ahead of Bay at 73.5 and Harmonic Security at 71.5. Scores run 0 to 100 on weights fixed before any vendor was scored. ## The ranking for this situation | Rank | Tool | Total | Sees (20) | Stops (20) | Deployment (15) | Beyond (25) | Policy (10) | Transparency (5) | Proof (5) | | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | | 1 | [Microsoft](/platforms/microsoft/) | 75.5 | 8 | 8 | 6 | 8 | 9 | 7 | 4 | | 2 | [Bay](/platforms/bay/) | 73.5 | 8 | 8 | 9 | 7 | 6 | 1 | 8 | | 3 | [Harmonic Security](/platforms/harmonic-security/) | 71.5 | 8 | 7 | 7 | 7 | 7 | 6 | 7 | | 4 | [Island](/platforms/island/) | 69.5 | 7 | 7 | 6 | 8 | 7 | 3 | 8 | | 5 | [Aim Security](/platforms/aim-security/) | 68.5 | 8 | 7 | 5 | 7 | 6 | 7 | 8 | | 6 | [CrowdStrike](/platforms/crowdstrike/) | 66.0 | 8 | 7 | 6 | 6 | 7 | 3 | 7 | | 7 | [Zenity](/platforms/zenity/) | 65.5 | 6 | 7 | 5 | 7 | 7 | 7 | 8 | | 8 | [SentinelOne](/platforms/sentinelone/) | 63.5 | 7 | 7 | 5 | 7 | 6 | 4 | 5 | | 9 | [Noma Security](/platforms/noma-security/) | 63.5 | 8 | 8 | 6 | 5 | 5 | 4 | 6 | | 10 | [WitnessAI](/platforms/witnessai/) | 61.5 | 8 | 6 | 5 | 6 | 6 | 3 | 7 | | 11 | [Knostic](/platforms/knostic/) | 58.0 | 7 | 7 | 7 | 3 | 6 | 7 | 5 | | 12 | [Lasso Security](/platforms/lasso-security/) | 53.0 | 6 | 5 | 5 | 5 | 5 | 4 | 8 | | 13 | [LayerX](/platforms/layerx/) | 50.5 | 5 | 5 | 5 | 5 | 6 | 4 | 5 | | 14 | [Koi](/platforms/koi/) | 48.5 | 7 | 5 | 6 | 2 | 4 | 6 | 7 | | 15 | [Palo Alto Networks](/platforms/palo-alto-networks/) | 48.0 | 3 | 5 | 4 | 6 | 5 | 4 | 8 | | 16 | [Pillar Security](/platforms/pillar-security/) | 48.0 | 6 | 6 | 6 | 2 | 5 | 4 | 6 | All 16 tools on this situation's weights. | Tied at | Tools in order | Beyond (heaviest) | What sets the order | | --- | --- | --- | --- | | 63.5 | SentinelOne, Noma Security | 7, 5 | the higher Beyond score | | 48.0 | Palo Alto Networks, Pillar Security | 6, 2 | the higher Beyond score | Tied totals break on the higher score on the heaviest criterion, then by name. ## The weights for this situation Staff-wide assistants and browser agents live in the browser and in SaaS as much as on the laptop, so coverage beyond the endpoint under one policy weighs most. | Criterion | This situation | Site index | | --- | --- | --- | | Sees agentic activity on the endpoint | 20 | 25 | | Stops the action | 20 | 25 | | Deployment friction | 15 | 15 | | Beyond the endpoint | 25 | 10 | | Policy and governance | 10 | 10 | | Transparency | 5 | 10 | | Proof | 5 | 5 | Fixed 12 Nov 2026 before any score. ## Verdict Microsoft leads at 75.5 of 100, ahead of Bay at 73.5 and Harmonic Security at 71.5. Microsoft leads on Policy (9 to 6), Transparency (7 to 1) and Beyond (8 to 7), while Bay is ahead on Deployment (9 to 6) and Proof (named customers, funding, analyst coverage) (8 to 4). Bay ranks 2nd of 16 at 73.5, behind Microsoft at 75.5. Microsoft's lead over Bay comes from policy and governance (9 to 6) and transparency (7 to 1). ## Strength of the top three - [Microsoft](/platforms/microsoft/), 75.5: Defender discovers dozens of local agents and MCP servers on Windows and macOS (GA) and can block prompt-injection-driven actions in Claude Code, Codex CLI and GitHub Copilot CLI on Windows, with Endpoint DLP blocking pastes and uploads to AI sites. - [Bay](/platforms/bay/), 73.5: Bay documents on-device, session-aware allow/ask/deny/steer enforcement with named action types, agentless MDM/EDR deployment in under 5 minutes (Bay's claim), and one policy across endpoint, browser and cloud. - [Harmonic Security](/platforms/harmonic-security/), 71.5: Harmonic documents hook-based inline blocking of prompts and tool calls across named coding agents (Claude Code, Codex, Cursor, Copilot CLI and others) plus an MCP gateway with identity-aware access groups. ## Who should pick what - Pick: Microsoft at 75.5. - Runner-up: Bay at 73.5. Other buyer situations: [all five, with their leaders](/best-for/). ## What would change this ranking Microsoft is 2.0 points clear of Bay for company-wide AI rollouts. The smallest single change that would put Bay first is a Beyond score of 8 instead of 7 (weight 25). Scores move only when a sourced fact changes ([how](/method/)). ## Sources - learn.microsoft.com/en-us/defender-endpoint/local-agent-discovery-overview: https://learn.microsoft.com/en-us/defender-endpoint/local-agent-discovery-overview (read 12 Nov 2026) - learn.microsoft.com/en-us/defender-endpoint/ai-agent-support-matrix: https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-support-matrix (read 12 Nov 2026) - learn.microsoft.com/en-us/defender-endpoint/whats-new-in-microsoft-defender-endpoint: https://learn.microsoft.com/en-us/defender-endpoint/whats-new-in-microsoft-defender-endpoint (read 12 Nov 2026) - softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control: https://softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control (read 12 Nov 2026) - siliconangle.com/2026/11/10/ai-security-startup-bay-launches-with-50m-to-vet-agent-actions-before-they-run: https://siliconangle.com/2026/11/10/ai-security-startup-bay-launches-with-50m-to-vet-agent-actions-before-they-run/ (read 12 Nov 2026) - docs.harmonicsecurity.app/portal-guides/integration-guides/configure-opentelemetry: https://docs.harmonicsecurity.app/portal-guides/integration-guides/configure-opentelemetry (read 12 Nov 2026) - docs.harmonicsecurity.app/portal-guides/integration-guides/configure-anthropic-compliance-api-connector: https://docs.harmonicsecurity.app/portal-guides/integration-guides/configure-anthropic-compliance-api-connector (read 12 Nov 2026) - docs.harmonicsecurity.app/protect-agent/installation-and-deployment/frequently-asked-questions: https://docs.harmonicsecurity.app/protect-agent/installation-and-deployment/frequently-asked-questions (read 12 Nov 2026) - docs.harmonicsecurity.app/portal-guides/detection-and-response/sensitive-data-controls-for-coding-agents-and-desktop-applications: https://docs.harmonicsecurity.app/portal-guides/detection-and-response/sensitive-data-controls-for-coding-agents-and-desktop-applications (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/best-for/staff-rollout/ --- # Bay leads for a 30-day board inventory, 77.5 to Harmonic Security's 73.5 BEST FOR, read 13 Nov 2026 **The short answer.** For a CISO asked by the board for an inventory of AI agents within 30 days, Bay ranks first of 16 at 77.5 of 100, ahead of Harmonic Security at 73.5 and Microsoft at 72.0. Scores run 0 to 100 on weights fixed before any vendor was scored. ## The ranking for this situation | Rank | Tool | Total | Sees (40) | Stops (5) | Deployment (30) | Beyond (10) | Policy (5) | Transparency (5) | Proof (5) | | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | | 1 | [Bay](/platforms/bay/) | 77.5 | 8 | 8 | 9 | 7 | 6 | 1 | 8 | | 2 | [Harmonic Security](/platforms/harmonic-security/) | 73.5 | 8 | 7 | 7 | 7 | 7 | 6 | 7 | | 3 | [Microsoft](/platforms/microsoft/) | 72.0 | 8 | 8 | 6 | 8 | 9 | 7 | 4 | | 4 | [Aim Security](/platforms/aim-security/) | 68.0 | 8 | 7 | 5 | 7 | 6 | 7 | 8 | | 5 | [CrowdStrike](/platforms/crowdstrike/) | 68.0 | 8 | 7 | 6 | 6 | 7 | 3 | 7 | | 6 | [Noma Security](/platforms/noma-security/) | 66.5 | 8 | 8 | 6 | 5 | 5 | 4 | 6 | | 7 | [Island](/platforms/island/) | 66.5 | 7 | 7 | 6 | 8 | 7 | 3 | 8 | | 8 | [Knostic](/platforms/knostic/) | 64.5 | 7 | 7 | 7 | 3 | 6 | 7 | 5 | | 9 | [WitnessAI](/platforms/witnessai/) | 64.0 | 8 | 6 | 5 | 6 | 6 | 3 | 7 | | 10 | [SentinelOne](/platforms/sentinelone/) | 61.0 | 7 | 7 | 5 | 7 | 6 | 4 | 5 | | 11 | [Zenity](/platforms/zenity/) | 60.5 | 6 | 7 | 5 | 7 | 7 | 7 | 8 | | 12 | [Koi](/platforms/koi/) | 59.0 | 7 | 5 | 6 | 2 | 4 | 6 | 7 | | 13 | [Lasso Security](/platforms/lasso-security/) | 55.0 | 6 | 5 | 5 | 5 | 5 | 4 | 8 | | 14 | [Pillar Security](/platforms/pillar-security/) | 54.5 | 6 | 6 | 6 | 2 | 5 | 4 | 6 | | 15 | [LayerX](/platforms/layerx/) | 50.0 | 5 | 5 | 5 | 5 | 6 | 4 | 5 | | 16 | [Palo Alto Networks](/platforms/palo-alto-networks/) | 41.0 | 3 | 5 | 4 | 6 | 5 | 4 | 8 | All 16 tools on this situation's weights. | Tied at | Tools in order | Sees (heaviest) | What sets the order | | --- | --- | --- | --- | | 68.0 | Aim Security, CrowdStrike | 8, 8 | name order, as the scores are equal | | 66.5 | Noma Security, Island | 8, 7 | the higher Sees score | Tied totals break on the higher score on the heaviest criterion, then by name. ## The weights for this situation A 30-day inventory deadline rewards the widest view of agents on the most machines, deployed fastest; blocking can follow later. | Criterion | This situation | Site index | | --- | --- | --- | | Sees agentic activity on the endpoint | 40 | 25 | | Stops the action | 5 | 25 | | Deployment friction | 30 | 15 | | Beyond the endpoint | 10 | 10 | | Policy and governance | 5 | 10 | | Transparency | 5 | 10 | | Proof | 5 | 5 | Fixed 12 Nov 2026 before any score. ## Verdict Bay leads at 77.5 of 100, ahead of Harmonic Security at 73.5 and Microsoft at 72.0. Bay leads on Deployment (9 to 7), Stops (8 to 7) and Proof (named customers, funding, analyst coverage) (8 to 7), while Harmonic Security is ahead on Transparency (6 to 1) and Policy (7 to 6). Bay ranks first here at 77.5, ahead of Harmonic Security at 73.5. Bay is a young company that left stealth in 2026, publishes no price and has little independent testing behind its claims. ## Strength of the top three - [Bay](/platforms/bay/), 77.5: Bay documents on-device, session-aware allow/ask/deny/steer enforcement with named action types, agentless MDM/EDR deployment in under 5 minutes (Bay's claim), and one policy across endpoint, browser and cloud. - [Harmonic Security](/platforms/harmonic-security/), 73.5: Harmonic documents hook-based inline blocking of prompts and tool calls across named coding agents (Claude Code, Codex, Cursor, Copilot CLI and others) plus an MCP gateway with identity-aware access groups. - [Microsoft](/platforms/microsoft/), 72.0: Defender discovers dozens of local agents and MCP servers on Windows and macOS (GA) and can block prompt-injection-driven actions in Claude Code, Codex CLI and GitHub Copilot CLI on Windows, with Endpoint DLP blocking pastes and uploads to AI sites. ## Who should pick what - Pick: Bay at 77.5. - Runner-up: Harmonic Security at 73.5. Other buyer situations: [all five, with their leaders](/best-for/). ## What would change this ranking Bay is 4.0 points clear of Harmonic Security for a 30-day board inventory. The smallest single change that would put Harmonic Security first is a Deployment score of 9 instead of 7 (weight 30). Scores move only when a sourced fact changes ([how](/method/)). ## Sources - softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control: https://softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control (read 12 Nov 2026) - siliconangle.com/2026/11/10/ai-security-startup-bay-launches-with-50m-to-vet-agent-actions-before-they-run: https://siliconangle.com/2026/11/10/ai-security-startup-bay-launches-with-50m-to-vet-agent-actions-before-they-run/ (read 12 Nov 2026) - docs.harmonicsecurity.app/portal-guides/integration-guides/configure-opentelemetry: https://docs.harmonicsecurity.app/portal-guides/integration-guides/configure-opentelemetry (read 12 Nov 2026) - docs.harmonicsecurity.app/portal-guides/integration-guides/configure-anthropic-compliance-api-connector: https://docs.harmonicsecurity.app/portal-guides/integration-guides/configure-anthropic-compliance-api-connector (read 12 Nov 2026) - docs.harmonicsecurity.app/protect-agent/installation-and-deployment/frequently-asked-questions: https://docs.harmonicsecurity.app/protect-agent/installation-and-deployment/frequently-asked-questions (read 12 Nov 2026) - docs.harmonicsecurity.app/portal-guides/detection-and-response/sensitive-data-controls-for-coding-agents-and-desktop-applications: https://docs.harmonicsecurity.app/portal-guides/detection-and-response/sensitive-data-controls-for-coding-agents-and-desktop-applications (read 12 Nov 2026) - learn.microsoft.com/en-us/defender-endpoint/local-agent-discovery-overview: https://learn.microsoft.com/en-us/defender-endpoint/local-agent-discovery-overview (read 12 Nov 2026) - learn.microsoft.com/en-us/defender-endpoint/ai-agent-support-matrix: https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-support-matrix (read 12 Nov 2026) - learn.microsoft.com/en-us/defender-endpoint/whats-new-in-microsoft-defender-endpoint: https://learn.microsoft.com/en-us/defender-endpoint/whats-new-in-microsoft-defender-endpoint (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/best-for/board-30-days/ --- # Five buyer situations, five rankings, and the leader of each BEST FOR, read 13 Nov 2026 **The short answer.** The index is re-ranked five times on weights written for a buyer situation: coding agents (Bay and Microsoft tied at 75.5, Bay first on the tie-break), regulated data (Microsoft), lean team with EDR (Bay), staff rollout (Microsoft), board in 30 days (Bay). Each weight set was fixed before any score and is published on its page. ## Who should pick what - [Coding agents](/best-for/coding-agents/) (an engineering organization where every developer runs coding agents with shell access): pick Bay at 75.5, level with Microsoft and first on the tie-break. - [Regulated data](/best-for/regulated-data/) (a regulated company that must show which AI tools touch customer data): pick Microsoft at 79.0, with Harmonic Security next at 72.5. Bay ranks 3rd of 16 at 70.5. - [Lean team with EDR](/best-for/lean-team-edr/) (a lean security team with EDR in place and no view of AI agents): pick Bay at 76.5, with Microsoft next at 73.5. - [Staff rollout](/best-for/staff-rollout/) (a company rolling out an AI assistant and browser agents to all staff): pick Microsoft at 75.5, with Bay next at 73.5. Bay ranks 2nd of 16 at 73.5. - [Board in 30 days](/best-for/board-30-days/) (a CISO asked by the board for an inventory of AI agents within 30 days): pick Bay at 77.5, with Harmonic Security next at 73.5. ## What would change this ranking Microsoft is 3.5 points clear of Bay on the published weights. The smallest single change that would put Bay first is a Policy score of 10 instead of 6 (weight 10). Scores move only when a sourced fact changes ([how](/method/)). ## Top three in each situation | Situation | First | Second | Third | | --- | --- | --- | --- | | [Coding agents](/best-for/coding-agents/) | Bay 75.5 (tied) | Microsoft 75.5 (tied) | Harmonic Security 72.0 | | [Regulated data](/best-for/regulated-data/) | Microsoft 79.0 | Harmonic Security 72.5 | Bay 70.5 | | [Lean team with EDR](/best-for/lean-team-edr/) | Bay 76.5 | Microsoft 73.5 | Harmonic Security 72.5 | | [Staff rollout](/best-for/staff-rollout/) | Microsoft 75.5 | Bay 73.5 | Harmonic Security 71.5 | | [Board in 30 days](/best-for/board-30-days/) | Bay 77.5 | Harmonic Security 73.5 | Microsoft 72.0 | Top three per situation, total out of 100. **About Bay.** Bay is a young company that left stealth in 2026, publishes no price and has little independent testing behind its claims. ## Sources - Method: https://aiendpointsecurityranked.com/method/ (read 13 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/best-for/ --- # Microsoft, Harmonic Security and SentinelOne see the most agent surfaces: 5 of 6 documented each DATASET, read 13 Nov 2026 **The short answer.** The Agent Surface Map covers 16 tools and 6 surfaces, 96 cells in all: 60 documented, 26 partial and 10 none found on 13 Nov 2026. Microsoft, Harmonic Security and SentinelOne have the most documented cells, 5 of 6 each. Rows follow the site ranking. ## The Agent Surface Map as a table Rows follow the site ranking. Cells read documented, partial or none found on 13 Nov 2026. | Tool | Documented | Desktop apps | IDE assistants | Command-line agents | MCP servers | Browser extensions | SaaS copilots | | --- | --- | --- | --- | --- | --- | --- | --- | | [Microsoft](/agent-surface-map/microsoft/) | 5 of 6 | documented | documented | documented | documented | partial | documented | | [Bay](/agent-surface-map/bay/) | 4 of 6 | partial | documented | documented | documented | documented | partial | | [Harmonic Security](/agent-surface-map/harmonic-security/) | 5 of 6 | documented | documented | documented | documented | documented | partial | | [Aim Security](/agent-surface-map/aim-security/) | 4 of 6 | documented | partial | documented | documented | documented | partial | | [CrowdStrike](/agent-surface-map/crowdstrike/) | 4 of 6 | documented | documented | documented | documented | partial | partial | | [Noma Security](/agent-surface-map/noma-security/) | 4 of 6 | partial | documented | documented | documented | none found | documented | | [Island](/agent-surface-map/island/) | 4 of 6 | documented | documented | partial | documented | documented | partial | | [Zenity](/agent-surface-map/zenity/) | 4 of 6 | partial | documented | documented | documented | partial | documented | | [Knostic](/agent-surface-map/knostic/) | 4 of 6 | documented | documented | documented | documented | partial | partial | | [SentinelOne](/agent-surface-map/sentinelone/) | 5 of 6 | documented | documented | documented | documented | documented | partial | | [WitnessAI](/agent-surface-map/witnessai/) | 3 of 6 | documented | documented | partial | documented | partial | partial | | [Koi](/agent-surface-map/koi/) | 4 of 6 | partial | documented | documented | documented | documented | none found | | [Lasso Security](/agent-surface-map/lasso-security/) | 3 of 6 | partial | documented | documented | documented | none found | partial | | [Pillar Security](/agent-surface-map/pillar-security/) | 3 of 6 | none found | documented | documented | documented | none found | none found | | [LayerX](/agent-surface-map/layerx/) | 2 of 6 | documented | none found | none found | none found | documented | partial | | [Palo Alto Networks](/agent-surface-map/palo-alto-networks/) | 2 of 6 | partial | documented | partial | none found | documented | partial | Agent Surface Map, edition of 13 Nov 2026. Open a tool's row for each cell's note, source and date. ## Verdict by situation - **Coding agents:** Bay and Microsoft tie at 75.5 and both score 8 on Stops, the heaviest criterion here, so name order puts Bay 1st. Bay has 4 of 6 surfaces documented; among the top three, Microsoft and Harmonic Security document the most, 5 of 6 each. - **Regulated data:** Microsoft ranks first at 79.0 with 5 of 6 surfaces documented; among the top three, Microsoft and Harmonic Security document the most, 5 of 6 each. Bay ranks 3rd of 16 at 70.5, behind Microsoft at 79.0. Microsoft's lead over Bay comes from policy and governance (9 to 6) and transparency (7 to 1). - **Lean team with EDR:** Bay ranks first at 76.5 with 4 of 6 surfaces documented; among the top three, Microsoft and Harmonic Security document the most, 5 of 6 each. - **Staff rollout:** Microsoft ranks first at 75.5 with 5 of 6 surfaces documented; among the top three, Microsoft and Harmonic Security document the most, 5 of 6 each. Bay ranks 2nd of 16 at 73.5, behind Microsoft at 75.5. Microsoft's lead over Bay comes from policy and governance (9 to 6) and transparency (7 to 1). - **Board in 30 days:** Bay ranks first at 77.5 with 4 of 6 surfaces documented; among the top three, Harmonic Security and Microsoft document the most, 5 of 6 each. Bay is a young company that left stealth in 2026, publishes no price and has little independent testing behind its claims. Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). ## By surface - [Desktop apps](/surfaces/desktop-apps/): 9 of 16 tools documented - [IDE assistants](/surfaces/ide-assistants/): 14 of 16 tools documented - [Command-line agents](/surfaces/cli-agents/): 12 of 16 tools documented - [MCP servers](/surfaces/mcp-servers/): 14 of 16 tools documented - [Browser extensions](/surfaces/browser-extensions/): 8 of 16 tools documented - [SaaS copilots](/surfaces/saas-copilots/): 3 of 16 tools documented ## Download and cite CSV and JSON under CC BY 4.0: [agent-surface-map.csv](/datasets/agent-surface-map.csv) and [agent-surface-map.json](/datasets/agent-surface-map.json). The [methodology](/agent-surface-map/methodology/) explains how a cell is read. Cite this: AI Endpoint Security Ranked, Agent Surface Map, Action Control Matrix and scores, edition of 13 Nov 2026, aiendpointsecurityranked.com/datasets/, CC BY 4.0. ## Sources - Microsoft: learn.microsoft.com/en-us/defender-endpoint/ai-agent-support-matrix: https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-support-matrix (read 12 Nov 2026) - Microsoft: learn.microsoft.com/en-us/microsoft-365/admin/manage/agent-shadow-ai: https://learn.microsoft.com/en-us/microsoft-365/admin/manage/agent-shadow-ai (read 12 Nov 2026) - Microsoft: learn.microsoft.com/en-us/defender-endpoint/ai-agent-runtime-protection-overview: https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-runtime-protection-overview (read 12 Nov 2026) - Microsoft: learn.microsoft.com/en-us/defender-endpoint/local-agent-discovery-overview: https://learn.microsoft.com/en-us/defender-endpoint/local-agent-discovery-overview (read 12 Nov 2026) - Microsoft: learn.microsoft.com/en-us/purview/dspm-for-ai: https://learn.microsoft.com/en-us/purview/dspm-for-ai (read 12 Nov 2026) - Microsoft: learn.microsoft.com/en-us/purview/endpoint-dlp-learn-about: https://learn.microsoft.com/en-us/purview/endpoint-dlp-learn-about (read 12 Nov 2026) - Microsoft: learn.microsoft.com/en-us/purview/ai-microsoft-purview: https://learn.microsoft.com/en-us/purview/ai-microsoft-purview (read 12 Nov 2026) - Microsoft: www.microsoft.com/en-us/security/blog/2026/05/01/microsoft-agent-365-now-generally-available-expands-capabilities-and-integrations: https://www.microsoft.com/en-us/security/blog/2026/05/01/microsoft-agent-365-now-generally-available-expands-capabilities-and-integrations/ (read 12 Nov 2026) - Bay: www.bay.io/platform/ai-endpoint-security: https://www.bay.io/platform/ai-endpoint-security (read 12 Nov 2026) - Bay: softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control: https://softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control (read 12 Nov 2026) - Bay: www.bay.io/platform/cloud-workspaces: https://www.bay.io/platform/cloud-workspaces (read 12 Nov 2026) - Bay: www.bay.io/platform: https://www.bay.io/platform (read 12 Nov 2026) - Bay: www.bay.io/platform/browser: https://www.bay.io/platform/browser (read 12 Nov 2026) - Bay: www.bay.io/blog/bay-emerges-from-stealth-with-50m-after-fortune-100-companies-used-its-platform-to-triple-autonomous-ai-agent-use-2: https://www.bay.io/blog/bay-emerges-from-stealth-with-50m-after-fortune-100-companies-used-its-platform-to-triple-autonomous-ai-agent-use-2 (read 12 Nov 2026) - Harmonic Security: www.harmonic.security/solutions/endpoint-ai-security: https://www.harmonic.security/solutions/endpoint-ai-security (read 12 Nov 2026) - Harmonic Security: docs.harmonicsecurity.app/portal-guides/detection-and-response/sensitive-data-controls-for-coding-agents-and-desktop-applications: https://docs.harmonicsecurity.app/portal-guides/detection-and-response/sensitive-data-controls-for-coding-agents-and-desktop-applications (read 12 Nov 2026) - Harmonic Security: www.harmonic.security/pricing: https://www.harmonic.security/pricing (read 12 Nov 2026) - Harmonic Security: docs.harmonicsecurity.app/portal-guides/managing-your-organization/configure-mcp-server-access-controls: https://docs.harmonicsecurity.app/portal-guides/managing-your-organization/configure-mcp-server-access-controls (read 12 Nov 2026) - Harmonic Security: docs.harmonicsecurity.app/portal-guides/detection-and-response/configure-scenarios-and-interventions: https://docs.harmonicsecurity.app/portal-guides/detection-and-response/configure-scenarios-and-interventions (read 12 Nov 2026) - Harmonic Security: www.harmonic.security/products/command: https://www.harmonic.security/products/command (read 12 Nov 2026) - Harmonic Security: docs.harmonicsecurity.app/llms.txt: https://docs.harmonicsecurity.app/llms.txt (read 12 Nov 2026) - Harmonic Security: docs.harmonicsecurity.app/mcp-gateway/getting-started/what-is-mcp-gateway: https://docs.harmonicsecurity.app/mcp-gateway/getting-started/what-is-mcp-gateway (read 12 Nov 2026) - Harmonic Security: docs.harmonicsecurity.app/portal-guides/integration-guides/configure-anthropic-compliance-api-connector: https://docs.harmonicsecurity.app/portal-guides/integration-guides/configure-anthropic-compliance-api-connector (read 12 Nov 2026) - Aim Security: www.catonetworks.com/platform/cato-ai-endpoint: https://www.catonetworks.com/platform/cato-ai-endpoint/ (read 12 Nov 2026) - Aim Security: www.catonetworks.com/news/cato-acquires-aim-security-to-extend-sase-leadership-and-secure-enterprise-ai-transformation: https://www.catonetworks.com/news/cato-acquires-aim-security-to-extend-sase-leadership-and-secure-enterprise-ai-transformation/ (read 12 Nov 2026) - Aim Security: www.catonetworks.com/platform/agentic-ai-security: https://www.catonetworks.com/platform/agentic-ai-security/ (read 12 Nov 2026) - CrowdStrike: www.msspalert.com/news/crowdstrike-brings-ai-security-to-the-endpoint-and-the-timing-makes-sense: https://www.msspalert.com/news/crowdstrike-brings-ai-security-to-the-endpoint-and-the-timing-makes-sense (read 12 Nov 2026) - CrowdStrike: www.crowdstrike.com/en-us/press-releases/crowdstrike-announces-general-availability-of-falcon-ai-detection-and-response: https://www.crowdstrike.com/en-us/press-releases/crowdstrike-announces-general-availability-of-falcon-ai-detection-and-response/ (read 12 Nov 2026) - CrowdStrike: www.crowdstrike.com/en-us/blog/crowdstrike-advances-ai-security-and-safety-falcon-guardian: https://www.crowdstrike.com/en-us/blog/crowdstrike-advances-ai-security-and-safety-falcon-guardian/ (read 12 Nov 2026) - CrowdStrike: www.crowdstrike.com/en-us/blog/crowdstrike-shadow-AI-visibility-service: https://www.crowdstrike.com/en-us/blog/crowdstrike-shadow-AI-visibility-service/ (read 12 Nov 2026) - CrowdStrike: www.crowdstrike.com/en-us/blog/falcon-guardian-defines-next-generation-of-ai-security: https://www.crowdstrike.com/en-us/blog/falcon-guardian-defines-next-generation-of-ai-security/ (read 12 Nov 2026) - CrowdStrike: www.crowdstrike.com/en-us/platform/falcon-guardian-aidr: https://www.crowdstrike.com/en-us/platform/falcon-guardian-aidr/ (read 12 Nov 2026) - CrowdStrike: www.crowdstrike.com/en-us/blog/crowdstrike-brings-enterprise-browser-security-to-falcon-platform: https://www.crowdstrike.com/en-us/blog/crowdstrike-brings-enterprise-browser-security-to-falcon-platform/ (read 12 Nov 2026) - CrowdStrike: www.crowdstrike.com/en-us/blog/crowdstrike-extends-endpoint-security-to-stop-supply-chain-attacks: https://www.crowdstrike.com/en-us/blog/crowdstrike-extends-endpoint-security-to-stop-supply-chain-attacks/ (read 12 Nov 2026) - Noma Security: noma.security/platform/endpoint-agents: https://noma.security/platform/endpoint-agents (read 12 Nov 2026) - Noma Security: noma.security/blog/noma-brings-agent-security-to-every-employee-endpoint: https://noma.security/blog/noma-brings-agent-security-to-every-employee-endpoint (read 12 Nov 2026) - Noma Security: noma.security/blog/securing-the-agentic-frontier-noma-unveils-the-first-real-time-agent-runtime-security-for-cursor: https://noma.security/blog/securing-the-agentic-frontier-noma-unveils-the-first-real-time-agent-runtime-security-for-cursor (read 12 Nov 2026) - Noma Security: github.com/Noma-Security/noma-marketplace: https://github.com/Noma-Security/noma-marketplace (read 12 Nov 2026) - Noma Security: noma.security/solutions/mcp-security: https://noma.security/solutions/mcp-security (read 12 Nov 2026) - Noma Security: noma.security: https://noma.security/ (read 12 Nov 2026) - Noma Security: github.com/Noma-Security/noma-marketplace/blob/main/guardrails/README.md: https://github.com/Noma-Security/noma-marketplace/blob/main/guardrails/README.md (read 12 Nov 2026) - Island: www.island.io/enterprise-desktop: https://www.island.io/enterprise-desktop (read 12 Nov 2026) - Island: www.island.io/product/ai: https://www.island.io/product/ai (read 12 Nov 2026) - Island: www.island.io/blog/expanding-island-to-govern-your-ai-workforce: https://www.island.io/blog/expanding-island-to-govern-your-ai-workforce (read 12 Nov 2026) - Island: www.island.io/blog/claude-code-cowork-and-claude-for-microsoft-365-now-report-into-island: https://www.island.io/blog/claude-code-cowork-and-claude-for-microsoft-365-now-report-into-island (read 12 Nov 2026) - Island: www.island.io/blog/see-score-and-control-every-ai-asset-on-your-endpoints: https://www.island.io/blog/see-score-and-control-every-ai-asset-on-your-endpoints (read 12 Nov 2026) - Island: www.island.io/product/enterprise-browser-extension: https://www.island.io/product/enterprise-browser-extension (read 12 Nov 2026) - Island: www.island.io/press/island-brings-the-enterprise-browser-to-ai-agents: https://www.island.io/press/island-brings-the-enterprise-browser-to-ai-agents (read 12 Nov 2026) - Island: www.island.io/blog/ai-runtime-protection-governing-what-agents-do-not-just-what-theyre-told: https://www.island.io/blog/ai-runtime-protection-governing-what-agents-do-not-just-what-theyre-told (read 12 Nov 2026) - Island: www.securityweek.com/island-raises-400-million-at-6-4-billion-valuation: https://www.securityweek.com/island-raises-400-million-at-6-4-billion-valuation/ (read 12 Nov 2026) - Island: betanews.com/2025/12/17/governing-ai-where-work-actually-happens-qa: https://betanews.com/2025/12/17/governing-ai-where-work-actually-happens-qa/ (read 12 Nov 2026) - Zenity: zenity.io/use-cases/agent-type/coding-personal-agents: https://zenity.io/use-cases/agent-type/coding-personal-agents (read 12 Nov 2026) - Zenity: zenity.io/blog/coding-agent-attack-surface: https://zenity.io/blog/coding-agent-attack-surface (read 12 Nov 2026) - Zenity: zenity.io/platform/mcp-security: https://zenity.io/platform/mcp-security (read 12 Nov 2026) - Zenity: zenity.io/press-release/zenity-raises-125-million-to-secure-the-era-of-1-billion-ai-agents: https://zenity.io/press-release/zenity-raises-125-million-to-secure-the-era-of-1-billion-ai-agents (read 12 Nov 2026) - Zenity: zenity.io/company: https://zenity.io/company (read 12 Nov 2026) - Zenity: zenity.io: https://zenity.io/ (read 12 Nov 2026) - Zenity: zenity.io/press-release/zenity-extends-ai-agent-security-and-governance-to-claude-enterprise: https://zenity.io/press-release/zenity-extends-ai-agent-security-and-governance-to-claude-enterprise (read 12 Nov 2026) - Zenity: zenity.io/platform/ai-detection-and-response: https://zenity.io/platform/ai-detection-and-response (read 12 Nov 2026) - Knostic: www.knostic.ai/faq: https://www.knostic.ai/faq (read 12 Nov 2026) - Knostic: www.knostic.ai/kirin: https://www.knostic.ai/kirin (read 12 Nov 2026) - Knostic: www.knostic.ai/pricing: https://www.knostic.ai/pricing (read 12 Nov 2026) - Knostic: siliconangle.com/2025/03/05/knostic-raises-11m-strengthen-enterprise-ai-security-need-know-access-controls: https://siliconangle.com/2025/03/05/knostic-raises-11m-strengthen-enterprise-ai-security-need-know-access-controls/ (read 12 Nov 2026) - Knostic: www.knostic.ai: https://www.knostic.ai (read 12 Nov 2026) - SentinelOne: www.sentinelone.com/blog/a-new-chapter-for-ai-and-cybersecurity-sentinelone-acquires-prompt-security: https://www.sentinelone.com/blog/a-new-chapter-for-ai-and-cybersecurity-sentinelone-acquires-prompt-security/ (read 12 Nov 2026) - SentinelOne: www.sentinelone.com/press/sentinelone-expands-ai-security-coverage-to-claude-cowork: https://www.sentinelone.com/press/sentinelone-expands-ai-security-coverage-to-claude-cowork/ (read 12 Nov 2026) - SentinelOne: www.sentinelone.com/blog/prompt-security-for-agentic-ai: https://www.sentinelone.com/blog/prompt-security-for-agentic-ai/ (read 12 Nov 2026) - SentinelOne: www.sentinelone.com/platform/prompt-security-ai-usage-control: https://www.sentinelone.com/platform/prompt-security-ai-usage-control/ (read 13 Nov 2026) - SentinelOne: www.sentinelone.com/press/sentinelone-launches-singularity-agentic-detection-and-response-to-stop-rogue-and-malicious-ai-agents-at-machine-speed: https://www.sentinelone.com/press/sentinelone-launches-singularity-agentic-detection-and-response-to-stop-rogue-and-malicious-ai-agents-at-machine-speed/ (read 12 Nov 2026) - SentinelOne: siliconangle.com/2026/03/23/sentinelone-snyk-introduce-new-tools-securing-ai-agents: https://siliconangle.com/2026/03/23/sentinelone-snyk-introduce-new-tools-securing-ai-agents/ (read 12 Nov 2026) - SentinelOne: prompt.security/blog/the-key-layer-in-ai-security-browser-and-endpoint-sensors: https://prompt.security/blog/the-key-layer-in-ai-security-browser-and-endpoint-sensors (read 12 Nov 2026) - SentinelOne: chromewebstore.google.com/search/prompt%20security: https://chromewebstore.google.com/search/prompt%20security (read 12 Nov 2026) - SentinelOne: www.sentinelone.com/press/sentinelone-named-an-ai-security-platform-leader-in-latio-2026-ai-security-market-report: https://www.sentinelone.com/press/sentinelone-named-an-ai-security-platform-leader-in-latio-2026-ai-security-market-report/ (read 12 Nov 2026) - WitnessAI: witness.ai/for-agents: https://witness.ai/for-agents/ (read 12 Nov 2026) - WitnessAI: witness.ai/product: https://witness.ai/product/ (read 12 Nov 2026) - WitnessAI: witness.ai: https://witness.ai/ (read 12 Nov 2026) - WitnessAI: witness.ai/for-employees: https://witness.ai/for-employees/ (read 12 Nov 2026) - WitnessAI: witness.ai/blog/witnessai-in-sacrs-arise-market-map-why-a-unified-ai-control-plane-matters: https://witness.ai/blog/witnessai-in-sacrs-arise-market-map-why-a-unified-ai-control-plane-matters/ (read 12 Nov 2026) - WitnessAI: witness.ai/protect: https://witness.ai/protect/ (read 12 Nov 2026) - Koi: www.paloaltonetworks.com/resources/techbriefs/koi-agentic-endpoint-security: https://www.paloaltonetworks.com/resources/techbriefs/koi-agentic-endpoint-security (read 12 Nov 2026) - Koi: www.paloaltonetworks.com/resources/datasheets/cortex-agentic-endpoint-security-at-a-glance: https://www.paloaltonetworks.com/resources/datasheets/cortex-agentic-endpoint-security-at-a-glance (read 12 Nov 2026) - Koi: www.paloaltonetworks.com/company/press/2026/palo-alto-networks-completes-acquisition-of-koi-to-secure-the-agentic-endpoint: https://www.paloaltonetworks.com/company/press/2026/palo-alto-networks-completes-acquisition-of-koi-to-secure-the-agentic-endpoint (read 12 Nov 2026) - Koi: www.paloaltonetworks.com/blog/2026/02/securing-the-agentic-endpoint: https://www.paloaltonetworks.com/blog/2026/02/securing-the-agentic-endpoint/ (read 12 Nov 2026) - Koi: www.sdxcentral.com/news/palo-alto-networks-bolsters-ai-security-for-the-endpoint: https://www.sdxcentral.com/news/palo-alto-networks-bolsters-ai-security-for-the-endpoint/ (read 13 Nov 2026) - Koi: cortex-docs.paloaltonetworks.com/cortex-xdr-6.x-rn/readme/features-introduced-in-2026/november-2026/release-highlights.md: https://cortex-docs.paloaltonetworks.com/cortex-xdr-6.x-rn/readme/features-introduced-in-2026/november-2026/release-highlights.md (read 13 Nov 2026) - Koi: www.paloaltonetworks.com/cortex/agentic-endpoint-security.md: https://www.paloaltonetworks.com/cortex/agentic-endpoint-security.md (read 12 Nov 2026) - Koi: www.calcalistech.com/ctechnews/article/xpz78znuc: https://www.calcalistech.com/ctechnews/article/xpz78znuc (read 12 Nov 2026) - Koi: www.securityweek.com/palo-alto-networks-to-acquire-koi-in-reported-400-million-transaction: https://www.securityweek.com/palo-alto-networks-to-acquire-koi-in-reported-400-million-transaction/ (read 12 Nov 2026) - Lasso Security: www.lasso.security/platform/ai-usage-control: https://www.lasso.security/platform/ai-usage-control (read 12 Nov 2026) - Lasso Security: www.lasso.security/blog/securing-ai-on-endpoints-how-to-govern-ai-tools-coding-agents-and-mcp-connections: https://www.lasso.security/blog/securing-ai-on-endpoints-how-to-govern-ai-tools-coding-agents-and-mcp-connections (read 12 Nov 2026) - Lasso Security: www.lasso.security/use-cases/ai-coding-assistants: https://www.lasso.security/use-cases/ai-coding-assistants (read 12 Nov 2026) - Lasso Security: www.lasso.security/use-cases/mcp-security: https://www.lasso.security/use-cases/mcp-security (read 12 Nov 2026) - Lasso Security: github.com/lasso-security/mcp-gateway: https://github.com/lasso-security/mcp-gateway (read 12 Nov 2026) - Lasso Security: www.lasso.security: https://www.lasso.security (read 12 Nov 2026) - Lasso Security: www.lasso.security/blog/bringing-claude-enterprise-activity-into-lasso-with-claudes-compliance-api: https://www.lasso.security/blog/bringing-claude-enterprise-activity-into-lasso-with-claudes-compliance-api (read 12 Nov 2026) - Lasso Security: github.com/lasso-security/kong-ai-lasso-guardrail: https://github.com/lasso-security/kong-ai-lasso-guardrail (read 12 Nov 2026) - Pillar Security: www.pillar.security/platform/agentic-endpoint: https://www.pillar.security/platform/agentic-endpoint (read 12 Nov 2026) - Pillar Security: pulse2.com/vanta-to-acquire-pillar-security-to-expand-ai-agent-protection-as-customer-base-surpasses-20000: https://pulse2.com/vanta-to-acquire-pillar-security-to-expand-ai-agent-protection-as-customer-base-surpasses-20000/ (read 12 Nov 2026) - LayerX: www.akamai.com/solutions/ai-usage-control: https://www.akamai.com/solutions/ai-usage-control (read 12 Nov 2026) - LayerX: www.manilatimes.net/2026/05/21/tmt-newswire/globenewswire/layerx-security-integrates-with-new-extended-plan-for-aws-security-hub/2348650: https://www.manilatimes.net/2026/05/21/tmt-newswire/globenewswire/layerx-security-integrates-with-new-extended-plan-for-aws-security-hub/2348650 (read 13 Nov 2026) - LayerX: www.akamai.com/products/akamai-workforce-protector: https://www.akamai.com/products/akamai-workforce-protector (read 12 Nov 2026) - Palo Alto Networks: www.paloaltonetworks.com/prisma/prisma-ai-runtime-security: https://www.paloaltonetworks.com/prisma/prisma-ai-runtime-security (read 12 Nov 2026) - Palo Alto Networks: www.paloaltonetworks.com/blog/2026/10/reinventing-endpoint-security-with-cortex-xdr-one: https://www.paloaltonetworks.com/blog/2026/10/reinventing-endpoint-security-with-cortex-xdr-one/ (read 12 Nov 2026) - Palo Alto Networks: www.paloaltonetworks.com/sase/prisma-browser: https://www.paloaltonetworks.com/sase/prisma-browser (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/agent-surface-map/ --- # A Agent Surface Map cell is yes only when public documentation names the surface METHODOLOGY, read 13 Nov 2026 **The short answer.** Each cell of the Agent Surface Map answers which agent surfaces it sees, as documented on 13 Nov 2026. Yes needs a documented capability that names the surface, partial covers a limited or conditional case, and none found means no public source named it. The matrix holds 96 cells across 16 tools. ## What the levels mean - **Yes:** The vendor documents inventory or control of agents on this surface, naming it. - **Partial:** The documentation names a limited case, a condition, a roadmap item or detection after the fact. - **None found:** No public source read on 13 Nov 2026 names it. This is written "none found on 13 Nov 2026" and is not a claim about the product. ## The 6 surfaces - **Desktop apps:** Chat and agent applications installed on a laptop, such as a desktop assistant that can read local folders or run tools on the user's behalf. - **IDE assistants:** Assistants inside code editors and IDEs, which read the open repository and can propose or apply edits and commands. - **Command-line agents:** Coding agents that run in a terminal with the user's shell, files and credentials, usually started by a developer. - **MCP servers:** Model Context Protocol servers running on or reached from the endpoint, which hand agents tools, files and data sources. - **Browser extensions:** AI extensions and agentic browsers that act on web pages, read form data and call model services from the browser. - **SaaS copilots:** Copilots and agents built into SaaS applications, which the endpoint only reaches through the browser or an API. ## How a cell is filled - Each cell cites the fact ids that support it in the vendor evidence file, with a quote, a source URL and a read date. - A vendor's own page counts as the vendor's claim and is labelled that way; a cell stays yes or partial on a vendor claim alone, while the criterion scores cap such claims at 7 of 10. - Cells were read on 13 Nov 2026. A cell changes only when a sourced fact changes, and the change is logged with the fact. - Rows are ordered by the site ranking, which the cells inform but do not decide: the Sees and Stops criteria also weigh depth and second sources. Across the 16 tools there are 60 documented cells, 26 partial cells and 10 cells where nothing was found on 13 Nov 2026. Cite this: AI Endpoint Security Ranked, Agent Surface Map, Action Control Matrix and scores, edition of 13 Nov 2026, aiendpointsecurityranked.com/datasets/, CC BY 4.0. ## Use the cells - [Open the Agent Surface Map](/agent-surface-map/) - [The six agent surfaces](/surfaces/) - [Download the datasets](/datasets/) - [The scoring method](/method/) ## Sources - Agent Surface Map: https://aiendpointsecurityranked.com/agent-surface-map/ (read 13 Nov 2026) - Scoring method: https://aiendpointsecurityranked.com/method/ (read 13 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/agent-surface-map/methodology/ --- # Microsoft sees 5 of 6 agent surfaces, with 1 more in part DATASET ROW, read 13 Nov 2026 **The short answer.** Microsoft ranks 1st of 16 on the index, and on the Agent Surface Map it has 5 documented cells, 1 partial and 0 none found on 13 Nov 2026. Each cell below carries a note, a source and a read date. Rows follow the site ranking. Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). ## Microsoft, cell by cell Desktop apps documented; IDE assistants documented; Command-line agents documented; MCP servers documented; Browser extensions partial; SaaS copilots documented. | Surface | Level | Note | Source and read date | | --- | --- | --- | --- | | Desktop apps | documented | Discovers Claude Desktop, ChatGPT Desktop, Codex Desktop and others; network inspection covers ChatGPT Desktop. | [learn.microsoft.com](https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-support-matrix) (12 Nov 2026); [learn.microsoft.com](https://learn.microsoft.com/en-us/microsoft-365/admin/manage/agent-shadow-ai) (12 Nov 2026) | | IDE assistants | documented | Discovers Cursor, Windsurf and VS Code assistant extensions; hooks cover GitHub Copilot extension. | [learn.microsoft.com](https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-support-matrix) (12 Nov 2026); [learn.microsoft.com](https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-support-matrix) (12 Nov 2026) | | Command-line agents | documented | Discovers Claude Code, Codex CLI, Gemini CLI; runtime protection covers Claude Code and Copilot CLI. | [learn.microsoft.com](https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-support-matrix) (12 Nov 2026); [learn.microsoft.com](https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-runtime-protection-overview) (12 Nov 2026) | | MCP servers | documented | Discovers configured local and remote MCP servers per agent. | [learn.microsoft.com](https://learn.microsoft.com/en-us/defender-endpoint/local-agent-discovery-overview) (12 Nov 2026); [learn.microsoft.com](https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-support-matrix) (12 Nov 2026) | | Browser extensions | partial | Purview browser extension and Endpoint DLP cover AI sites in browsers; no agent browser extension inventory found. | [learn.microsoft.com](https://learn.microsoft.com/en-us/purview/dspm-for-ai) (12 Nov 2026); [learn.microsoft.com](https://learn.microsoft.com/en-us/purview/endpoint-dlp-learn-about) (12 Nov 2026) | | SaaS copilots | documented | Purview and Agent 365 cover Microsoft 365 Copilot, Copilot Studio and enterprise AI apps. | [learn.microsoft.com](https://learn.microsoft.com/en-us/purview/ai-microsoft-purview) (12 Nov 2026); [microsoft.com](https://www.microsoft.com/en-us/security/blog/2026/05/01/microsoft-agent-365-now-generally-available-expands-capabilities-and-integrations/) (12 Nov 2026) | Microsoft on the Agent Surface Map. Microsoft: Defender discovers dozens of local agents and MCP servers on Windows and macOS (GA) and can block prompt-injection-driven actions in Claude Code, Codex CLI and GitHub Copilot CLI on Windows, with Endpoint DLP blocking pastes and uploads to AI sites. Rows for every tool: [the Agent Surface Map](/agent-surface-map/). Method: [how a cell is read](/agent-surface-map/methodology/). All [six agent surfaces](/surfaces/), the [full index](/ranking/) and [every head-to-head](/compare/). ## Sources - learn.microsoft.com/en-us/defender-endpoint/ai-agent-support-matrix: https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-support-matrix (read 12 Nov 2026) - learn.microsoft.com/en-us/microsoft-365/admin/manage/agent-shadow-ai: https://learn.microsoft.com/en-us/microsoft-365/admin/manage/agent-shadow-ai (read 12 Nov 2026) - learn.microsoft.com/en-us/defender-endpoint/ai-agent-runtime-protection-overview: https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-runtime-protection-overview (read 12 Nov 2026) - learn.microsoft.com/en-us/defender-endpoint/local-agent-discovery-overview: https://learn.microsoft.com/en-us/defender-endpoint/local-agent-discovery-overview (read 12 Nov 2026) - learn.microsoft.com/en-us/purview/dspm-for-ai: https://learn.microsoft.com/en-us/purview/dspm-for-ai (read 12 Nov 2026) - learn.microsoft.com/en-us/purview/endpoint-dlp-learn-about: https://learn.microsoft.com/en-us/purview/endpoint-dlp-learn-about (read 12 Nov 2026) - learn.microsoft.com/en-us/purview/ai-microsoft-purview: https://learn.microsoft.com/en-us/purview/ai-microsoft-purview (read 12 Nov 2026) - www.microsoft.com/en-us/security/blog/2026/05/01/microsoft-agent-365-now-generally-available-expands-capabilities-and-integrations: https://www.microsoft.com/en-us/security/blog/2026/05/01/microsoft-agent-365-now-generally-available-expands-capabilities-and-integrations/ (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/agent-surface-map/microsoft/ --- # Bay sees 4 of 6 agent surfaces, with 2 more in part DATASET ROW, read 13 Nov 2026 **The short answer.** Bay ranks 2nd of 16 on the index, and on the Agent Surface Map it has 4 documented cells, 2 partial and 0 none found on 13 Nov 2026. Each cell below carries a note, a source and a read date. Rows follow the site ranking. ## Bay, cell by cell Desktop apps partial; IDE assistants documented; Command-line agents documented; MCP servers documented; Browser extensions documented; SaaS copilots partial. | Surface | Level | Note | Source and read date | | --- | --- | --- | --- | | Desktop apps | partial | Bay says it finds agents and models on employee devices and names Claude Code and Cursor; desktop chat apps are not named on its pages. | [bay.io](https://www.bay.io/platform/ai-endpoint-security?utm_source=aiendpointsecurityranked.com&utm_medium=referral&utm_campaign=asset) (12 Nov 2026); [bay.io](https://www.bay.io/platform/ai-endpoint-security?utm_source=aiendpointsecurityranked.com&utm_medium=referral&utm_campaign=asset) (12 Nov 2026) | | IDE assistants | documented | Bay names Cursor, and Software Analyst notes IDE extensions in the entity graph. | [bay.io](https://www.bay.io/platform/ai-endpoint-security?utm_source=aiendpointsecurityranked.com&utm_medium=referral&utm_campaign=asset) (12 Nov 2026); [softwareanalyst.substack.com](https://softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control) (12 Nov 2026) | | Command-line agents | documented | Bay names Claude Code, Codex, Gemini CLI and GitHub Copilot, and lists CLI tools in its inventory. | [bay.io](https://www.bay.io/platform/ai-endpoint-security?utm_source=aiendpointsecurityranked.com&utm_medium=referral&utm_campaign=asset) (12 Nov 2026); [bay.io](https://www.bay.io/platform/ai-endpoint-security?utm_source=aiendpointsecurityranked.com&utm_medium=referral&utm_campaign=asset) (12 Nov 2026) | | MCP servers | documented | Bay inventories MCP servers and can scope which servers may read or write. | [bay.io](https://www.bay.io/platform/ai-endpoint-security?utm_source=aiendpointsecurityranked.com&utm_medium=referral&utm_campaign=asset) (12 Nov 2026); [bay.io](https://www.bay.io/platform?utm_source=aiendpointsecurityranked.com&utm_medium=referral&utm_campaign=asset) (12 Nov 2026) | | Browser extensions | documented | Bay finds browser AI tools, extensions and embedded SaaS agents and can block domains and extensions. | [bay.io](https://www.bay.io/platform/browser?utm_source=aiendpointsecurityranked.com&utm_medium=referral&utm_campaign=asset) (12 Nov 2026); [bay.io](https://www.bay.io/platform/browser?utm_source=aiendpointsecurityranked.com&utm_medium=referral&utm_campaign=asset) (12 Nov 2026) | | SaaS copilots | partial | Bay says it surfaces embedded SaaS agents in the browser and covers SaaS applications; no named SaaS copilot list found. | [bay.io](https://www.bay.io/platform/browser?utm_source=aiendpointsecurityranked.com&utm_medium=referral&utm_campaign=asset) (12 Nov 2026); [bay.io](https://www.bay.io/blog/bay-emerges-from-stealth-with-50m-after-fortune-100-companies-used-its-platform-to-triple-autonomous-ai-agent-use-2?utm_source=aiendpointsecurityranked.com&utm_medium=referral&utm_campaign=asset) (12 Nov 2026) | Bay on the Agent Surface Map. Bay: Bay documents on-device, session-aware allow/ask/deny/steer enforcement with named action types, agentless MDM/EDR deployment in under 5 minutes (Bay's claim), and one policy across endpoint, browser and cloud. Rows for every tool: [the Agent Surface Map](/agent-surface-map/). Method: [how a cell is read](/agent-surface-map/methodology/). **About Bay.** Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). All [six agent surfaces](/surfaces/), the [full index](/ranking/) and [every head-to-head](/compare/). ## Sources - www.bay.io/platform/ai-endpoint-security: https://www.bay.io/platform/ai-endpoint-security (read 12 Nov 2026) - softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control: https://softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control (read 12 Nov 2026) - www.bay.io/platform/cloud-workspaces: https://www.bay.io/platform/cloud-workspaces (read 12 Nov 2026) - www.bay.io/platform: https://www.bay.io/platform (read 12 Nov 2026) - www.bay.io/platform/browser: https://www.bay.io/platform/browser (read 12 Nov 2026) - www.bay.io/blog/bay-emerges-from-stealth-with-50m-after-fortune-100-companies-used-its-platform-to-triple-autonomous-ai-agent-use-2: https://www.bay.io/blog/bay-emerges-from-stealth-with-50m-after-fortune-100-companies-used-its-platform-to-triple-autonomous-ai-agent-use-2 (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/agent-surface-map/bay/ --- # Harmonic Security sees 5 of 6 agent surfaces, with 1 more in part DATASET ROW, read 13 Nov 2026 **The short answer.** Harmonic Security ranks 3rd of 16 on the index, and on the Agent Surface Map it has 5 documented cells, 1 partial and 0 none found on 13 Nov 2026. Each cell below carries a note, a source and a read date. Rows follow the site ranking. Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). ## Harmonic Security, cell by cell Desktop apps documented; IDE assistants documented; Command-line agents documented; MCP servers documented; Browser extensions documented; SaaS copilots partial. | Surface | Level | Note | Source and read date | | --- | --- | --- | --- | | Desktop apps | documented | Discovery names Claude Desktop and ChatGPT Desktop; inline prompt controls are early access and Windows only. | [harmonic.security](https://www.harmonic.security/solutions/endpoint-ai-security) (12 Nov 2026); [docs.harmonicsecurity.app](https://docs.harmonicsecurity.app/portal-guides/detection-and-response/sensitive-data-controls-for-coding-agents-and-desktop-applications) (12 Nov 2026) | | IDE assistants | documented | Cursor, VS Code and Copilot covered by discovery and hook-based prompt and tool-call controls. | [harmonic.security](https://www.harmonic.security/solutions/endpoint-ai-security) (12 Nov 2026); [harmonic.security](https://www.harmonic.security/solutions/endpoint-ai-security) (12 Nov 2026) | | Command-line agents | documented | Claude Code, Codex CLI, Copilot CLI, Kiro CLI hooks documented; Grok Build blocks tool calls only. | [harmonic.security](https://www.harmonic.security/solutions/endpoint-ai-security) (12 Nov 2026); [docs.harmonicsecurity.app](https://docs.harmonicsecurity.app/portal-guides/detection-and-response/sensitive-data-controls-for-coding-agents-and-desktop-applications) (12 Nov 2026) | | MCP servers | documented | Discovery plus local MCP gateway with block on unpermitted servers and tool controls. | [harmonic.security](https://www.harmonic.security/pricing) (12 Nov 2026); [docs.harmonicsecurity.app](https://docs.harmonicsecurity.app/portal-guides/managing-your-organization/configure-mcp-server-access-controls) (12 Nov 2026) | | Browser extensions | documented | Browser extension deployed alongside the Endpoint Agent; scope of AI browser agents not documented. | [harmonic.security](https://www.harmonic.security/products/command) (12 Nov 2026) | | SaaS copilots | partial | Claude for Microsoft 365 activity and Microsoft Copilot in Teams visibility documented; cloud MCP gateway early access. | [docs.harmonicsecurity.app](https://docs.harmonicsecurity.app/llms.txt) (12 Nov 2026); [docs.harmonicsecurity.app](https://docs.harmonicsecurity.app/mcp-gateway/getting-started/what-is-mcp-gateway) (12 Nov 2026) | Harmonic Security on the Agent Surface Map. Harmonic Security: Harmonic documents hook-based inline blocking of prompts and tool calls across named coding agents (Claude Code, Codex, Cursor, Copilot CLI and others) plus an MCP gateway with identity-aware access groups. Rows for every tool: [the Agent Surface Map](/agent-surface-map/). Method: [how a cell is read](/agent-surface-map/methodology/). All [six agent surfaces](/surfaces/), the [full index](/ranking/) and [every head-to-head](/compare/). ## Sources - www.harmonic.security/solutions/endpoint-ai-security: https://www.harmonic.security/solutions/endpoint-ai-security (read 12 Nov 2026) - docs.harmonicsecurity.app/portal-guides/detection-and-response/sensitive-data-controls-for-coding-agents-and-desktop-applications: https://docs.harmonicsecurity.app/portal-guides/detection-and-response/sensitive-data-controls-for-coding-agents-and-desktop-applications (read 12 Nov 2026) - www.harmonic.security/pricing: https://www.harmonic.security/pricing (read 12 Nov 2026) - docs.harmonicsecurity.app/portal-guides/managing-your-organization/configure-mcp-server-access-controls: https://docs.harmonicsecurity.app/portal-guides/managing-your-organization/configure-mcp-server-access-controls (read 12 Nov 2026) - docs.harmonicsecurity.app/portal-guides/detection-and-response/configure-scenarios-and-interventions: https://docs.harmonicsecurity.app/portal-guides/detection-and-response/configure-scenarios-and-interventions (read 12 Nov 2026) - www.harmonic.security/products/command: https://www.harmonic.security/products/command (read 12 Nov 2026) - docs.harmonicsecurity.app/llms.txt: https://docs.harmonicsecurity.app/llms.txt (read 12 Nov 2026) - docs.harmonicsecurity.app/mcp-gateway/getting-started/what-is-mcp-gateway: https://docs.harmonicsecurity.app/mcp-gateway/getting-started/what-is-mcp-gateway (read 12 Nov 2026) - docs.harmonicsecurity.app/portal-guides/integration-guides/configure-anthropic-compliance-api-connector: https://docs.harmonicsecurity.app/portal-guides/integration-guides/configure-anthropic-compliance-api-connector (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/agent-surface-map/harmonic-security/ --- # Aim Security sees 4 of 6 agent surfaces, with 2 more in part DATASET ROW, read 13 Nov 2026 **The short answer.** Aim Security ranks 4th of 16 on the index, and on the Agent Surface Map it has 4 documented cells, 2 partial and 0 none found on 13 Nov 2026. Each cell below carries a note, a source and a read date. Rows follow the site ranking. Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). ## Aim Security, cell by cell Desktop apps documented; IDE assistants partial; Command-line agents documented; MCP servers documented; Browser extensions documented; SaaS copilots partial. | Surface | Level | Note | Source and read date | | --- | --- | --- | --- | | Desktop apps | documented | Desktop AI apps are covered by the endpoint client and prompts are checked at send. | [catonetworks.com](https://www.catonetworks.com/platform/cato-ai-endpoint/) (12 Nov 2026); [catonetworks.com](https://www.catonetworks.com/platform/cato-ai-endpoint/) (12 Nov 2026) | | IDE assistants | partial | Cursor is named as a covered coding agent; no IDE-extension inventory documented. | [catonetworks.com](https://www.catonetworks.com/platform/cato-ai-endpoint/) (12 Nov 2026); [catonetworks.com](https://www.catonetworks.com/news/cato-acquires-aim-security-to-extend-sase-leadership-and-secure-enterprise-ai-transformation/) (12 Nov 2026) | | Command-line agents | documented | Local agents such as Claude Code and Codex are shown in the endpoint dashboard with MCP servers and skills. | [catonetworks.com](https://www.catonetworks.com/platform/cato-ai-endpoint/) (12 Nov 2026); [catonetworks.com](https://www.catonetworks.com/platform/cato-ai-endpoint/) (12 Nov 2026) | | MCP servers | documented | Finds MCP servers and skills per agent and profiles MCP servers and tool calls. | [catonetworks.com](https://www.catonetworks.com/platform/cato-ai-endpoint/) (12 Nov 2026); [catonetworks.com](https://www.catonetworks.com/platform/cato-ai-endpoint/) (12 Nov 2026) | | Browser extensions | documented | Browser AI is covered by the same client and inline inspection; a Cato browser extension is named by a customer. | [catonetworks.com](https://www.catonetworks.com/platform/cato-ai-endpoint/) (12 Nov 2026); [catonetworks.com](https://www.catonetworks.com/platform/cato-ai-endpoint/) (12 Nov 2026) | | SaaS copilots | partial | Microsoft Copilot use is named as securable through inline inspection; no SaaS-hosted agent controls documented in detail. | [catonetworks.com](https://www.catonetworks.com/news/cato-acquires-aim-security-to-extend-sase-leadership-and-secure-enterprise-ai-transformation/) (12 Nov 2026) | Aim Security on the Agent Surface Map. Aim Security: Cato AI Endpoint documents a lightweight client that finds local agents with their MCP servers and skills and blocks an unapproved tool call before it runs, in one dashboard with inline network inspection. Rows for every tool: [the Agent Surface Map](/agent-surface-map/). Method: [how a cell is read](/agent-surface-map/methodology/). All [six agent surfaces](/surfaces/), the [full index](/ranking/) and [every head-to-head](/compare/). ## Sources - www.catonetworks.com/platform/cato-ai-endpoint: https://www.catonetworks.com/platform/cato-ai-endpoint/ (read 12 Nov 2026) - www.catonetworks.com/news/cato-acquires-aim-security-to-extend-sase-leadership-and-secure-enterprise-ai-transformation: https://www.catonetworks.com/news/cato-acquires-aim-security-to-extend-sase-leadership-and-secure-enterprise-ai-transformation/ (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/agent-surface-map/aim-security/ --- # CrowdStrike sees 4 of 6 agent surfaces, with 2 more in part DATASET ROW, read 13 Nov 2026 **The short answer.** CrowdStrike ranks 5th of 16 on the index, and on the Agent Surface Map it has 4 documented cells, 2 partial and 0 none found on 13 Nov 2026. Each cell below carries a note, a source and a read date. Rows follow the site ranking. Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). ## CrowdStrike, cell by cell Desktop apps documented; IDE assistants documented; Command-line agents documented; MCP servers documented; Browser extensions partial; SaaS copilots partial. | Surface | Level | Note | Source and read date | | --- | --- | --- | --- | | Desktop apps | documented | A third party reports AIDR for Endpoint covers desktop AI apps; Guardian discovers agents on endpoints and blocks sensitive data in prompts. | [msspalert.com](https://www.msspalert.com/news/crowdstrike-brings-ai-security-to-the-endpoint-and-the-timing-makes-sense) (12 Nov 2026); [crowdstrike.com](https://www.crowdstrike.com/en-us/press-releases/crowdstrike-announces-general-availability-of-falcon-ai-detection-and-response/) (12 Nov 2026) | | IDE assistants | documented | Hooks cover Cursor and GitHub Copilot; discovery lists IDEs. | [crowdstrike.com](https://www.crowdstrike.com/en-us/blog/crowdstrike-advances-ai-security-and-safety-falcon-guardian/) (12 Nov 2026); [crowdstrike.com](https://www.crowdstrike.com/en-us/blog/crowdstrike-shadow-AI-visibility-service/) (12 Nov 2026) | | Command-line agents | documented | Hooks cover Claude Code, OpenAI Codex and OpenCode; Linux discovery for Claude Code and Codex is pre-beta. | [crowdstrike.com](https://www.crowdstrike.com/en-us/blog/crowdstrike-advances-ai-security-and-safety-falcon-guardian/) (12 Nov 2026); [crowdstrike.com](https://www.crowdstrike.com/en-us/blog/falcon-guardian-defines-next-generation-of-ai-security/) (12 Nov 2026) | | MCP servers | documented | Discovers MCP servers and malicious MCP servers and skills. | [crowdstrike.com](https://www.crowdstrike.com/en-us/platform/falcon-guardian-aidr/) (12 Nov 2026); [crowdstrike.com](https://www.crowdstrike.com/en-us/blog/crowdstrike-advances-ai-security-and-safety-falcon-guardian/) (12 Nov 2026) | | Browser extensions | partial | Extension inventory and blocking is in Falcon Seraphic Enterprise Browser, now on the Falcon platform. | [crowdstrike.com](https://www.crowdstrike.com/en-us/blog/crowdstrike-brings-enterprise-browser-security-to-falcon-platform/) (12 Nov 2026); [crowdstrike.com](https://www.crowdstrike.com/en-us/blog/crowdstrike-shadow-AI-visibility-service/) (12 Nov 2026) | | SaaS copilots | partial | Shadow SaaS agent discovery named for Copilot, Agentforce and ChatGPT Enterprise by a third party; Claude Cowork integration is pre-beta. | [msspalert.com](https://www.msspalert.com/news/crowdstrike-brings-ai-security-to-the-endpoint-and-the-timing-makes-sense) (12 Nov 2026); [crowdstrike.com](https://www.crowdstrike.com/en-us/blog/crowdstrike-advances-ai-security-and-safety-falcon-guardian/) (12 Nov 2026) | CrowdStrike on the Agent Surface Map. CrowdStrike: Falcon Guardian pairs OS-level sensor telemetry with agent-level hooks that can block actions before they run in Claude Code, Codex, GitHub Copilot, OpenCode and Cursor, plus package blocking in the same sensor. Rows for every tool: [the Agent Surface Map](/agent-surface-map/). Method: [how a cell is read](/agent-surface-map/methodology/). All [six agent surfaces](/surfaces/), the [full index](/ranking/) and [every head-to-head](/compare/). ## Sources - www.msspalert.com/news/crowdstrike-brings-ai-security-to-the-endpoint-and-the-timing-makes-sense: https://www.msspalert.com/news/crowdstrike-brings-ai-security-to-the-endpoint-and-the-timing-makes-sense (read 12 Nov 2026) - www.crowdstrike.com/en-us/press-releases/crowdstrike-announces-general-availability-of-falcon-ai-detection-and-response: https://www.crowdstrike.com/en-us/press-releases/crowdstrike-announces-general-availability-of-falcon-ai-detection-and-response/ (read 12 Nov 2026) - www.crowdstrike.com/en-us/blog/crowdstrike-advances-ai-security-and-safety-falcon-guardian: https://www.crowdstrike.com/en-us/blog/crowdstrike-advances-ai-security-and-safety-falcon-guardian/ (read 12 Nov 2026) - www.crowdstrike.com/en-us/blog/crowdstrike-shadow-AI-visibility-service: https://www.crowdstrike.com/en-us/blog/crowdstrike-shadow-AI-visibility-service/ (read 12 Nov 2026) - www.crowdstrike.com/en-us/blog/falcon-guardian-defines-next-generation-of-ai-security: https://www.crowdstrike.com/en-us/blog/falcon-guardian-defines-next-generation-of-ai-security/ (read 12 Nov 2026) - www.crowdstrike.com/en-us/platform/falcon-guardian-aidr: https://www.crowdstrike.com/en-us/platform/falcon-guardian-aidr/ (read 12 Nov 2026) - www.crowdstrike.com/en-us/blog/crowdstrike-brings-enterprise-browser-security-to-falcon-platform: https://www.crowdstrike.com/en-us/blog/crowdstrike-brings-enterprise-browser-security-to-falcon-platform/ (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/agent-surface-map/crowdstrike/ --- # Noma Security sees 4 of 6 agent surfaces, with 1 more in part DATASET ROW, read 13 Nov 2026 **The short answer.** Noma Security ranks 6th of 16 on the index, and on the Agent Surface Map it has 4 documented cells, 1 partial and 1 none found on 13 Nov 2026. Each cell below carries a note, a source and a read date. Rows follow the site ranking. Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). ## Noma Security, cell by cell Desktop apps partial; IDE assistants documented; Command-line agents documented; MCP servers documented; Browser extensions none found; SaaS copilots documented. | Surface | Level | Note | Source and read date | | --- | --- | --- | --- | | Desktop apps | partial | Claude Cowork named as an endpoint agent; desktop app inventory method not itemized. | [noma.security](https://noma.security/platform/endpoint-agents) (12 Nov 2026); [noma.security](https://noma.security/blog/noma-brings-agent-security-to-every-employee-endpoint) (12 Nov 2026) | | IDE assistants | documented | IDEs and Cursor, Windsurf, Kiro named; Cursor hooks integration documented. | [noma.security](https://noma.security/blog/noma-brings-agent-security-to-every-employee-endpoint) (12 Nov 2026); [noma.security](https://noma.security/blog/noma-brings-agent-security-to-every-employee-endpoint) (12 Nov 2026) | | Command-line agents | documented | Claude Code and Codex named on the endpoint page. | [noma.security](https://noma.security/platform/endpoint-agents) (12 Nov 2026); [noma.security](https://noma.security/blog/noma-brings-agent-security-to-every-employee-endpoint) (12 Nov 2026) | | MCP servers | documented | Approved and shadow MCP server discovery with tool exposure. | [noma.security](https://noma.security/solutions/mcp-security) (12 Nov 2026); [noma.security](https://noma.security/blog/noma-brings-agent-security-to-every-employee-endpoint) (12 Nov 2026) | | Browser extensions | none found | Nothing found on 12 Nov 2026 on browser extension or browser-agent controls. | none found on 12 Nov 2026 | | SaaS copilots | documented | Noma for SaaS Agents is a separate product page; Copilot Studio runtime guardrails posts exist. | [noma.security](https://noma.security/) (12 Nov 2026); [noma.security](https://noma.security/platform/endpoint-agents) (12 Nov 2026) | Noma Security on the Agent Surface Map. Noma Security: Has a dedicated, current endpoint page: discovery of agents, MCP servers and skills via existing EDR or MDM, plus action-level allow and block policy tied to IdP identity. Rows for every tool: [the Agent Surface Map](/agent-surface-map/). Method: [how a cell is read](/agent-surface-map/methodology/). All [six agent surfaces](/surfaces/), the [full index](/ranking/) and [every head-to-head](/compare/). ## Sources - noma.security/platform/endpoint-agents: https://noma.security/platform/endpoint-agents (read 12 Nov 2026) - noma.security/blog/noma-brings-agent-security-to-every-employee-endpoint: https://noma.security/blog/noma-brings-agent-security-to-every-employee-endpoint (read 12 Nov 2026) - noma.security/blog/securing-the-agentic-frontier-noma-unveils-the-first-real-time-agent-runtime-security-for-cursor: https://noma.security/blog/securing-the-agentic-frontier-noma-unveils-the-first-real-time-agent-runtime-security-for-cursor (read 12 Nov 2026) - github.com/Noma-Security/noma-marketplace: https://github.com/Noma-Security/noma-marketplace (read 12 Nov 2026) - noma.security/solutions/mcp-security: https://noma.security/solutions/mcp-security (read 12 Nov 2026) - noma.security: https://noma.security/ (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/agent-surface-map/noma-security/ --- # Island sees 4 of 6 agent surfaces, with 2 more in part DATASET ROW, read 13 Nov 2026 **The short answer.** Island ranks 7th of 16 on the index, and on the Agent Surface Map it has 4 documented cells, 2 partial and 0 none found on 13 Nov 2026. Each cell below carries a note, a source and a read date. Rows follow the site ranking. Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). ## Island, cell by cell Desktop apps documented; IDE assistants documented; Command-line agents partial; MCP servers documented; Browser extensions documented; SaaS copilots partial. | Surface | Level | Note | Source and read date | | --- | --- | --- | --- | | Desktop apps | documented | Island Desktop applies identity-aware policy and DLP to native apps and files on Windows and macOS; desktop agents are listed as governed. | [island.io](https://www.island.io/enterprise-desktop) (12 Nov 2026); [island.io](https://www.island.io/enterprise-desktop) (12 Nov 2026) | | IDE assistants | documented | Names Cursor, Windsurf and Cline and inventories IDE extensions and flags over-permissioned ones. | [island.io](https://www.island.io/product/ai) (12 Nov 2026); [island.io](https://www.island.io/blog/expanding-island-to-govern-your-ai-workforce) (12 Nov 2026) | | Command-line agents | partial | Claude Code sessions are read via the Compliance API and Bash commands are inspected; other CLI agents are not named. | [island.io](https://www.island.io/blog/claude-code-cowork-and-claude-for-microsoft-365-now-report-into-island) (12 Nov 2026); [island.io](https://www.island.io/product/ai) (12 Nov 2026) | | MCP servers | documented | Inventories MCP servers with risk scores and gateways MCP calls with policy and audit. | [island.io](https://www.island.io/product/ai) (12 Nov 2026); [island.io](https://www.island.io/product/ai) (12 Nov 2026) | | Browser extensions | documented | Own browser plus extension for Chrome, Edge, Firefox and Safari, and a browser for agents with per-agent policy. | [island.io](https://www.island.io/product/enterprise-browser-extension) (12 Nov 2026); [island.io](https://www.island.io/press/island-brings-the-enterprise-browser-to-ai-agents) (12 Nov 2026) | | SaaS copilots | partial | Claude for Microsoft 365 add-ins are read through the Compliance API; coverage of other SaaS copilots is described only in general terms. | [island.io](https://www.island.io/blog/claude-code-cowork-and-claude-for-microsoft-365-now-report-into-island) (12 Nov 2026) | Island on the Agent Surface Map. Island: Island combines an endpoint agent for Windows and macOS with a browser, extension and MCP gateway, and documents blocking of Bash execution, file reads and network connections as well as clipboard DLP. Rows for every tool: [the Agent Surface Map](/agent-surface-map/). Method: [how a cell is read](/agent-surface-map/methodology/). All [six agent surfaces](/surfaces/), the [full index](/ranking/) and [every head-to-head](/compare/). ## Sources - www.island.io/enterprise-desktop: https://www.island.io/enterprise-desktop (read 12 Nov 2026) - www.island.io/product/ai: https://www.island.io/product/ai (read 12 Nov 2026) - www.island.io/blog/expanding-island-to-govern-your-ai-workforce: https://www.island.io/blog/expanding-island-to-govern-your-ai-workforce (read 12 Nov 2026) - www.island.io/blog/claude-code-cowork-and-claude-for-microsoft-365-now-report-into-island: https://www.island.io/blog/claude-code-cowork-and-claude-for-microsoft-365-now-report-into-island (read 12 Nov 2026) - www.island.io/blog/see-score-and-control-every-ai-asset-on-your-endpoints: https://www.island.io/blog/see-score-and-control-every-ai-asset-on-your-endpoints (read 12 Nov 2026) - www.island.io/product/enterprise-browser-extension: https://www.island.io/product/enterprise-browser-extension (read 12 Nov 2026) - www.island.io/press/island-brings-the-enterprise-browser-to-ai-agents: https://www.island.io/press/island-brings-the-enterprise-browser-to-ai-agents (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/agent-surface-map/island/ --- # Zenity sees 4 of 6 agent surfaces, with 2 more in part DATASET ROW, read 13 Nov 2026 **The short answer.** Zenity ranks 8th of 16 on the index, and on the Agent Surface Map it has 4 documented cells, 2 partial and 0 none found on 13 Nov 2026. Each cell below carries a note, a source and a read date. Rows follow the site ranking. Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). ## Zenity, cell by cell Desktop apps partial; IDE assistants documented; Command-line agents documented; MCP servers documented; Browser extensions partial; SaaS copilots documented. | Surface | Level | Note | Source and read date | | --- | --- | --- | --- | | Desktop apps | partial | Cowork and Claude Enterprise desktop named on the coding and personal agents page; no inventory method detailed. | [zenity.io](https://zenity.io/use-cases/agent-type/coding-personal-agents) (12 Nov 2026) | | IDE assistants | documented | Cursor and GitHub Copilot named; configuration and hook telemetry documented. | [zenity.io](https://zenity.io/use-cases/agent-type/coding-personal-agents) (12 Nov 2026); [zenity.io](https://zenity.io/use-cases/agent-type/coding-personal-agents) (12 Nov 2026) | | Command-line agents | documented | Claude Code, Codex named with permission-mode and hook behavior. | [zenity.io](https://zenity.io/use-cases/agent-type/coding-personal-agents) (12 Nov 2026); [zenity.io](https://zenity.io/blog/coding-agent-attack-surface) (12 Nov 2026) | | MCP servers | documented | MCP inventory and tool-call governance, delivered via gateway and hooks. | [zenity.io](https://zenity.io/blog/coding-agent-attack-surface) (12 Nov 2026); [zenity.io](https://zenity.io/platform/mcp-security) (12 Nov 2026) | | Browser extensions | partial | Agentic browsers appear in the Series C quote and PleaseFix research, but no browser control product page was read. | [zenity.io](https://zenity.io/press-release/zenity-raises-125-million-to-secure-the-era-of-1-billion-ai-agents) (12 Nov 2026); [zenity.io](https://zenity.io/company) (12 Nov 2026) | | SaaS copilots | documented | Copilot Studio, Agentforce, ChatGPT Enterprise, Gemini covered; this is the core of the platform. | [zenity.io](https://zenity.io/) (12 Nov 2026); [zenity.io](https://zenity.io/press-release/zenity-raises-125-million-to-secure-the-era-of-1-billion-ai-agents) (12 Nov 2026) | Zenity on the Agent Surface Map. Zenity: Documents pre-execution blocking of coding-agent tool calls through native hooks and an MCP gateway, including credential-read-then-network sequences. Rows for every tool: [the Agent Surface Map](/agent-surface-map/). Method: [how a cell is read](/agent-surface-map/methodology/). All [six agent surfaces](/surfaces/), the [full index](/ranking/) and [every head-to-head](/compare/). ## Sources - zenity.io/use-cases/agent-type/coding-personal-agents: https://zenity.io/use-cases/agent-type/coding-personal-agents (read 12 Nov 2026) - zenity.io/blog/coding-agent-attack-surface: https://zenity.io/blog/coding-agent-attack-surface (read 12 Nov 2026) - zenity.io/platform/mcp-security: https://zenity.io/platform/mcp-security (read 12 Nov 2026) - zenity.io/press-release/zenity-raises-125-million-to-secure-the-era-of-1-billion-ai-agents: https://zenity.io/press-release/zenity-raises-125-million-to-secure-the-era-of-1-billion-ai-agents (read 12 Nov 2026) - zenity.io/company: https://zenity.io/company (read 12 Nov 2026) - zenity.io: https://zenity.io/ (read 12 Nov 2026) - zenity.io/press-release/zenity-extends-ai-agent-security-and-governance-to-claude-enterprise: https://zenity.io/press-release/zenity-extends-ai-agent-security-and-governance-to-claude-enterprise (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/agent-surface-map/zenity/ --- # Knostic sees 4 of 6 agent surfaces, with 2 more in part DATASET ROW, read 13 Nov 2026 **The short answer.** Knostic ranks 9th of 16 on the index, and on the Agent Surface Map it has 4 documented cells, 2 partial and 0 none found on 13 Nov 2026. Each cell below carries a note, a source and a read date. Rows follow the site ranking. Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). ## Knostic, cell by cell Desktop apps documented; IDE assistants documented; Command-line agents documented; MCP servers documented; Browser extensions partial; SaaS copilots partial. | Surface | Level | Note | Source and read date | | --- | --- | --- | --- | | Desktop apps | documented | Claude Desktop and Cowork supported via endpoint client and plugin. | [knostic.ai](https://www.knostic.ai/faq) (12 Nov 2026); [knostic.ai](https://www.knostic.ai/faq) (12 Nov 2026) | | IDE assistants | documented | Cursor, Windsurf, VS Code Copilot via IDE extension. | [knostic.ai](https://www.knostic.ai/faq) (12 Nov 2026); [knostic.ai](https://www.knostic.ai/kirin) (12 Nov 2026) | | Command-line agents | documented | Claude Code CLI supported; Codex and Gemini CLI listed on Kirin page, roadmap in FAQ. | [knostic.ai](https://www.knostic.ai/faq) (12 Nov 2026); [knostic.ai](https://www.knostic.ai/kirin) (12 Nov 2026) | | MCP servers | documented | MCP scanning, cross-org visibility and quarantine of risky servers. | [knostic.ai](https://www.knostic.ai/pricing) (12 Nov 2026); [knostic.ai](https://www.knostic.ai/faq) (12 Nov 2026) | | Browser extensions | partial | Browser chatbots seen via network traffic, visibility only; extensions in IDEs scanned. | [knostic.ai](https://www.knostic.ai/faq) (12 Nov 2026) | | SaaS copilots | partial | Earlier Knostic product covered Microsoft 365 Copilot; current Kirin scope is endpoint agents. | [siliconangle.com](https://siliconangle.com/2025/03/05/knostic-raises-11m-strengthen-enterprise-ai-security-need-know-access-controls/) (12 Nov 2026) | Knostic on the Agent Surface Map. Knostic: Knostic documents Enforce-mode blocking of secret file reads, risky MCP servers, destructive commands and unsafe packages inside agents, with a published $15 per user Team price and free tier. Rows for every tool: [the Agent Surface Map](/agent-surface-map/). Method: [how a cell is read](/agent-surface-map/methodology/). All [six agent surfaces](/surfaces/), the [full index](/ranking/) and [every head-to-head](/compare/). ## Sources - www.knostic.ai/faq: https://www.knostic.ai/faq (read 12 Nov 2026) - www.knostic.ai/kirin: https://www.knostic.ai/kirin (read 12 Nov 2026) - www.knostic.ai/pricing: https://www.knostic.ai/pricing (read 12 Nov 2026) - siliconangle.com/2025/03/05/knostic-raises-11m-strengthen-enterprise-ai-security-need-know-access-controls: https://siliconangle.com/2025/03/05/knostic-raises-11m-strengthen-enterprise-ai-security-need-know-access-controls/ (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/agent-surface-map/knostic/ --- # SentinelOne sees 5 of 6 agent surfaces, with 1 more in part DATASET ROW, read 13 Nov 2026 **The short answer.** SentinelOne ranks 10th of 16 on the index, and on the Agent Surface Map it has 5 documented cells, 1 partial and 0 none found on 13 Nov 2026. Each cell below carries a note, a source and a read date. Rows follow the site ranking. Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). ## SentinelOne, cell by cell Desktop apps documented; IDE assistants documented; Command-line agents documented; MCP servers documented; Browser extensions documented; SaaS copilots partial. | Surface | Level | Note | Source and read date | | --- | --- | --- | --- | | Desktop apps | documented | Agent discovers sanctioned and shadow AI on desktop apps; Claude Cowork VM agent announced. | [sentinelone.com](https://www.sentinelone.com/blog/a-new-chapter-for-ai-and-cybersecurity-sentinelone-acquires-prompt-security/) (12 Nov 2026); [sentinelone.com](https://www.sentinelone.com/press/sentinelone-expands-ai-security-coverage-to-claude-cowork/) (12 Nov 2026) | | IDE assistants | documented | Desktop IDEs and embedded agents in developer tools are named in discovery. | [sentinelone.com](https://www.sentinelone.com/blog/a-new-chapter-for-ai-and-cybersecurity-sentinelone-acquires-prompt-security/) (12 Nov 2026); [sentinelone.com](https://www.sentinelone.com/blog/prompt-security-for-agentic-ai/) (12 Nov 2026) | | Command-line agents | documented | Terminal-based assistants are named in discovery, and the AI Usage Control page names Claude Code as tracked. | [sentinelone.com](https://www.sentinelone.com/blog/a-new-chapter-for-ai-and-cybersecurity-sentinelone-acquires-prompt-security/) (12 Nov 2026); [sentinelone.com](https://www.sentinelone.com/blog/prompt-security-for-agentic-ai/) (12 Nov 2026) | | MCP servers | documented | MCP discovery plus a gateway with per-server risk scores. | [sentinelone.com](https://www.sentinelone.com/blog/prompt-security-for-agentic-ai/) (12 Nov 2026); [sentinelone.com](https://www.sentinelone.com/blog/a-new-chapter-for-ai-and-cybersecurity-sentinelone-acquires-prompt-security/) (12 Nov 2026) | | Browser extensions | documented | Browser extension for web AI and shadow AI discovery. | [sentinelone.com](https://www.sentinelone.com/blog/a-new-chapter-for-ai-and-cybersecurity-sentinelone-acquires-prompt-security/) (12 Nov 2026); [prompt.security](https://prompt.security/blog/the-key-layer-in-ai-security-browser-and-endpoint-sensors) (12 Nov 2026) | | SaaS copilots | partial | ADR (early access) extends inventory to Agentforce, Copilot Studio, ChatGPT workspace agents and Claude managed agents. | [sentinelone.com](https://www.sentinelone.com/press/sentinelone-named-an-ai-security-platform-leader-in-latio-2026-ai-security-market-report/) (12 Nov 2026); [sentinelone.com](https://www.sentinelone.com/press/sentinelone-named-an-ai-security-platform-leader-in-latio-2026-ai-security-market-report/) (12 Nov 2026) | SentinelOne on the Agent Surface Map. SentinelOne: SentinelOne documents an MCP gateway with allow, block, filter or redact actions and runtime tool-call blocking, plus a lightweight agent and extension that discover shadow AI across browsers, IDEs and terminal assistants. Rows for every tool: [the Agent Surface Map](/agent-surface-map/). Method: [how a cell is read](/agent-surface-map/methodology/). All [six agent surfaces](/surfaces/), the [full index](/ranking/) and [every head-to-head](/compare/). ## Sources - www.sentinelone.com/blog/a-new-chapter-for-ai-and-cybersecurity-sentinelone-acquires-prompt-security: https://www.sentinelone.com/blog/a-new-chapter-for-ai-and-cybersecurity-sentinelone-acquires-prompt-security/ (read 12 Nov 2026) - www.sentinelone.com/press/sentinelone-expands-ai-security-coverage-to-claude-cowork: https://www.sentinelone.com/press/sentinelone-expands-ai-security-coverage-to-claude-cowork/ (read 12 Nov 2026) - www.sentinelone.com/blog/prompt-security-for-agentic-ai: https://www.sentinelone.com/blog/prompt-security-for-agentic-ai/ (read 12 Nov 2026) - www.sentinelone.com/platform/prompt-security-ai-usage-control: https://www.sentinelone.com/platform/prompt-security-ai-usage-control/ (read 13 Nov 2026) - www.sentinelone.com/press/sentinelone-launches-singularity-agentic-detection-and-response-to-stop-rogue-and-malicious-ai-agents-at-machine-speed: https://www.sentinelone.com/press/sentinelone-launches-singularity-agentic-detection-and-response-to-stop-rogue-and-malicious-ai-agents-at-machine-speed/ (read 12 Nov 2026) - siliconangle.com/2026/03/23/sentinelone-snyk-introduce-new-tools-securing-ai-agents: https://siliconangle.com/2026/03/23/sentinelone-snyk-introduce-new-tools-securing-ai-agents/ (read 12 Nov 2026) - prompt.security/blog/the-key-layer-in-ai-security-browser-and-endpoint-sensors: https://prompt.security/blog/the-key-layer-in-ai-security-browser-and-endpoint-sensors (read 12 Nov 2026) - chromewebstore.google.com/search/prompt%20security: https://chromewebstore.google.com/search/prompt%20security (read 12 Nov 2026) - www.sentinelone.com/press/sentinelone-named-an-ai-security-platform-leader-in-latio-2026-ai-security-market-report: https://www.sentinelone.com/press/sentinelone-named-an-ai-security-platform-leader-in-latio-2026-ai-security-market-report/ (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/agent-surface-map/sentinelone/ --- # WitnessAI sees 3 of 6 agent surfaces, with 3 more in part DATASET ROW, read 13 Nov 2026 **The short answer.** WitnessAI ranks 11th of 16 on the index, and on the Agent Surface Map it has 3 documented cells, 3 partial and 0 none found on 13 Nov 2026. Each cell below carries a note, a source and a read date. Rows follow the site ranking. Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). ## WitnessAI, cell by cell Desktop apps documented; IDE assistants documented; Command-line agents partial; MCP servers documented; Browser extensions partial; SaaS copilots partial. | Surface | Level | Note | Source and read date | | --- | --- | --- | --- | | Desktop apps | documented | Says it covers desktop clients and native apps such as Windows 11 Copilot, via network traffic. | [witness.ai](https://witness.ai/for-agents/) (12 Nov 2026); [witness.ai](https://witness.ai/product/) (12 Nov 2026) | | IDE assistants | documented | Captures prompts and tool calls across supported Cursor, Claude Code and Codex workflows and IDEs. | [witness.ai](https://witness.ai/for-agents/) (12 Nov 2026); [witness.ai](https://witness.ai/for-agents/) (12 Nov 2026) | | Command-line agents | partial | Claude Code and Codex workflows are named as supported; coverage depends on traffic passing through the platform. | [witness.ai](https://witness.ai/for-agents/) (12 Nov 2026) | | MCP servers | documented | Catalogs MCP servers and the tools agents call. | [witness.ai](https://witness.ai/) (12 Nov 2026); [witness.ai](https://witness.ai/for-employees/) (12 Nov 2026) | | Browser extensions | partial | Sees agentic plugins employees install, but says it does not use browser extensions; no browser agent coverage detail found. | [witness.ai](https://witness.ai/for-employees/) (12 Nov 2026); [witness.ai](https://witness.ai/product/) (12 Nov 2026) | | SaaS copilots | partial | Names Windows 11 Copilot and Office 365; other SaaS copilots not named. | [witness.ai](https://witness.ai/product/) (12 Nov 2026) | WitnessAI on the Agent Surface Map. WitnessAI: WitnessAI documents inline blocking of unapproved MCP servers and tool calls across IDEs, chat apps and custom agents, with audit records per blocked call and one policy for humans and agents. Rows for every tool: [the Agent Surface Map](/agent-surface-map/). Method: [how a cell is read](/agent-surface-map/methodology/). All [six agent surfaces](/surfaces/), the [full index](/ranking/) and [every head-to-head](/compare/). ## Sources - witness.ai/for-agents: https://witness.ai/for-agents/ (read 12 Nov 2026) - witness.ai/product: https://witness.ai/product/ (read 12 Nov 2026) - witness.ai: https://witness.ai/ (read 12 Nov 2026) - witness.ai/for-employees: https://witness.ai/for-employees/ (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/agent-surface-map/witnessai/ --- # Koi sees 4 of 6 agent surfaces, with 1 more in part DATASET ROW, read 13 Nov 2026 **The short answer.** Koi ranks 12th of 16 on the index, and on the Agent Surface Map it has 4 documented cells, 1 partial and 1 none found on 13 Nov 2026. Each cell below carries a note, a source and a read date. Rows follow the site ranking. Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). ## Koi, cell by cell Desktop apps partial; IDE assistants documented; Command-line agents documented; MCP servers documented; Browser extensions documented; SaaS copilots none found. | Surface | Level | Note | Source and read date | | --- | --- | --- | --- | | Desktop apps | partial | Lists user-deployed AI applications as governed endpoint items, without naming ChatGPT or Claude desktop. | [paloaltonetworks.com](https://www.paloaltonetworks.com/resources/techbriefs/koi-agentic-endpoint-security) (12 Nov 2026) | | IDE assistants | documented | Discovers IDE plugins and extensions on endpoints. | [paloaltonetworks.com](https://www.paloaltonetworks.com/resources/datasheets/cortex-agentic-endpoint-security-at-a-glance) (12 Nov 2026); [paloaltonetworks.com](https://www.paloaltonetworks.com/resources/techbriefs/koi-agentic-endpoint-security) (12 Nov 2026) | | Command-line agents | documented | Names Claude Code and OpenClaw as the agent tools it addresses; no CLI agent inventory detail found. | [paloaltonetworks.com](https://www.paloaltonetworks.com/company/press/2026/palo-alto-networks-completes-acquisition-of-koi-to-secure-the-agentic-endpoint) (12 Nov 2026); [paloaltonetworks.com](https://www.paloaltonetworks.com/blog/2026/02/securing-the-agentic-endpoint/) (12 Nov 2026) | | MCP servers | documented | Discovers MCP servers on endpoints. | [paloaltonetworks.com](https://www.paloaltonetworks.com/resources/datasheets/cortex-agentic-endpoint-security-at-a-glance) (12 Nov 2026); [paloaltonetworks.com](https://www.paloaltonetworks.com/resources/techbriefs/koi-agentic-endpoint-security) (12 Nov 2026) | | Browser extensions | documented | Discovers browser extensions on endpoints. | [paloaltonetworks.com](https://www.paloaltonetworks.com/resources/datasheets/cortex-agentic-endpoint-security-at-a-glance) (12 Nov 2026); [paloaltonetworks.com](https://www.paloaltonetworks.com/resources/techbriefs/koi-agentic-endpoint-security) (12 Nov 2026) | | SaaS copilots | none found | No SaaS-hosted copilot coverage found in AES material. | none found on 12 Nov 2026 | Koi on the Agent Surface Map. Koi: Palo Alto Networks documents inventory of IDE plugins, browser extensions and MCP servers on endpoints, plus real-time blocking of risky downloads and agent actions, with standalone or Cortex XDR deployment. Rows for every tool: [the Agent Surface Map](/agent-surface-map/). Method: [how a cell is read](/agent-surface-map/methodology/). All [six agent surfaces](/surfaces/), the [full index](/ranking/) and [every head-to-head](/compare/). ## Sources - www.paloaltonetworks.com/resources/techbriefs/koi-agentic-endpoint-security: https://www.paloaltonetworks.com/resources/techbriefs/koi-agentic-endpoint-security (read 12 Nov 2026) - www.paloaltonetworks.com/resources/datasheets/cortex-agentic-endpoint-security-at-a-glance: https://www.paloaltonetworks.com/resources/datasheets/cortex-agentic-endpoint-security-at-a-glance (read 12 Nov 2026) - www.paloaltonetworks.com/company/press/2026/palo-alto-networks-completes-acquisition-of-koi-to-secure-the-agentic-endpoint: https://www.paloaltonetworks.com/company/press/2026/palo-alto-networks-completes-acquisition-of-koi-to-secure-the-agentic-endpoint (read 12 Nov 2026) - www.paloaltonetworks.com/blog/2026/02/securing-the-agentic-endpoint: https://www.paloaltonetworks.com/blog/2026/02/securing-the-agentic-endpoint/ (read 12 Nov 2026) - www.sdxcentral.com/news/palo-alto-networks-bolsters-ai-security-for-the-endpoint: https://www.sdxcentral.com/news/palo-alto-networks-bolsters-ai-security-for-the-endpoint/ (read 13 Nov 2026) - cortex-docs.paloaltonetworks.com/cortex-xdr-6.x-rn/readme/features-introduced-in-2026/november-2026/release-highlights.md: https://cortex-docs.paloaltonetworks.com/cortex-xdr-6.x-rn/readme/features-introduced-in-2026/november-2026/release-highlights.md (read 13 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/agent-surface-map/koi/ --- # Lasso Security sees 3 of 6 agent surfaces, with 2 more in part DATASET ROW, read 13 Nov 2026 **The short answer.** Lasso Security ranks 13th of 16 on the index, and on the Agent Surface Map it has 3 documented cells, 2 partial and 1 none found on 13 Nov 2026. Each cell below carries a note, a source and a read date. Rows follow the site ranking. Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). ## Lasso Security, cell by cell Desktop apps partial; IDE assistants documented; Command-line agents documented; MCP servers documented; Browser extensions none found; SaaS copilots partial. | Surface | Level | Note | Source and read date | | --- | --- | --- | --- | | Desktop apps | partial | Says it discovers local desktop agents; no named desktop apps or enforcement detail. | [lasso.security](https://www.lasso.security/platform/ai-usage-control) (12 Nov 2026); [lasso.security](https://www.lasso.security/blog/securing-ai-on-endpoints-how-to-govern-ai-tools-coding-agents-and-mcp-connections) (12 Nov 2026) | | IDE assistants | documented | Cursor and other coding assistants covered through hooks. | [lasso.security](https://www.lasso.security/use-cases/ai-coding-assistants) (12 Nov 2026); [lasso.security](https://www.lasso.security/use-cases/ai-coding-assistants) (12 Nov 2026) | | Command-line agents | documented | Claude Code, Codex and OpenCode covered through lifecycle hooks. | [lasso.security](https://www.lasso.security/use-cases/ai-coding-assistants) (12 Nov 2026); [lasso.security](https://www.lasso.security/use-cases/ai-coding-assistants) (12 Nov 2026) | | MCP servers | documented | MCP discovery, risk scoring and block at connection layer; open-source MCP gateway. | [lasso.security](https://www.lasso.security/use-cases/mcp-security) (12 Nov 2026); [lasso.security](https://www.lasso.security/use-cases/mcp-security) (12 Nov 2026) | | Browser extensions | none found | No browser extension or browser-agent inventory documented. | [lasso.security](https://www.lasso.security/platform/ai-usage-control) (12 Nov 2026) | | SaaS copilots | partial | SaaS chatbots and enterprise copilots mentioned in discovery; Claude Enterprise managed settings. | [lasso.security](https://www.lasso.security/platform/ai-usage-control) (12 Nov 2026); [lasso.security](https://www.lasso.security) (12 Nov 2026) | Lasso Security on the Agent Surface Map. Lasso Security: Lasso documents inspecting every coding-agent tool call before execution via Claude Code, Cursor and Codex hooks, with out-of-scope calls flagged or blocked, plus an open-source MCP gateway. Rows for every tool: [the Agent Surface Map](/agent-surface-map/). Method: [how a cell is read](/agent-surface-map/methodology/). All [six agent surfaces](/surfaces/), the [full index](/ranking/) and [every head-to-head](/compare/). ## Sources - www.lasso.security/platform/ai-usage-control: https://www.lasso.security/platform/ai-usage-control (read 12 Nov 2026) - www.lasso.security/blog/securing-ai-on-endpoints-how-to-govern-ai-tools-coding-agents-and-mcp-connections: https://www.lasso.security/blog/securing-ai-on-endpoints-how-to-govern-ai-tools-coding-agents-and-mcp-connections (read 12 Nov 2026) - www.lasso.security/use-cases/ai-coding-assistants: https://www.lasso.security/use-cases/ai-coding-assistants (read 12 Nov 2026) - www.lasso.security/use-cases/mcp-security: https://www.lasso.security/use-cases/mcp-security (read 12 Nov 2026) - github.com/lasso-security/mcp-gateway: https://github.com/lasso-security/mcp-gateway (read 12 Nov 2026) - www.lasso.security: https://www.lasso.security (read 12 Nov 2026) - www.lasso.security/blog/bringing-claude-enterprise-activity-into-lasso-with-claudes-compliance-api: https://www.lasso.security/blog/bringing-claude-enterprise-activity-into-lasso-with-claudes-compliance-api (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/agent-surface-map/lasso-security/ --- # Pillar Security sees 3 of 6 agent surfaces, with 0 more in part DATASET ROW, read 13 Nov 2026 **The short answer.** Pillar Security ranks 14th of 16 on the index, and on the Agent Surface Map it has 3 documented cells, 0 partial and 3 none found on 13 Nov 2026. Each cell below carries a note, a source and a read date. Rows follow the site ranking. Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). ## Pillar Security, cell by cell Desktop apps none found; IDE assistants documented; Command-line agents documented; MCP servers documented; Browser extensions none found; SaaS copilots none found. | Surface | Level | Note | Source and read date | | --- | --- | --- | --- | | Desktop apps | none found | Nothing found on 12 Nov 2026 for desktop AI apps such as ChatGPT or Claude desktop. | none found on 12 Nov 2026 | | IDE assistants | documented | Cursor deep scan and pre-action hook documented; Antigravity and others by fingerprint. | [pillar.security](https://www.pillar.security/platform/agentic-endpoint) (12 Nov 2026); [pillar.security](https://www.pillar.security/platform/agentic-endpoint) (12 Nov 2026) | | Command-line agents | documented | Claude Code, Codex CLI full runtime control; Gemini CLI, OpenCode and Pi scanned. | [pillar.security](https://www.pillar.security/platform/agentic-endpoint) (12 Nov 2026); [pillar.security](https://www.pillar.security/platform/agentic-endpoint) (12 Nov 2026) | | MCP servers | documented | Fleet MCP inventory, 23,000-server catalog risk score and per-server tool allow-lists. | [pillar.security](https://www.pillar.security/platform/agentic-endpoint) (12 Nov 2026); [pillar.security](https://www.pillar.security/platform/agentic-endpoint) (12 Nov 2026) | | Browser extensions | none found | Nothing found on 12 Nov 2026 on browser extensions or browser agents. | none found on 12 Nov 2026 | | SaaS copilots | none found | Nothing found on 12 Nov 2026 for SaaS-hosted copilots on the endpoint page. | none found on 12 Nov 2026 | Pillar Security on the Agent Surface Map. Pillar Security: Documents pre-action blocking in the agent's own pre-tool hook for Claude Code, Codex CLI and Cursor, with credential-exfiltration and persistence controls and MDM-installed hooks developers cannot disable. Rows for every tool: [the Agent Surface Map](/agent-surface-map/). Method: [how a cell is read](/agent-surface-map/methodology/). All [six agent surfaces](/surfaces/), the [full index](/ranking/) and [every head-to-head](/compare/). ## Sources - www.pillar.security/platform/agentic-endpoint: https://www.pillar.security/platform/agentic-endpoint (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/agent-surface-map/pillar-security/ --- # LayerX sees 2 of 6 agent surfaces, with 1 more in part DATASET ROW, read 13 Nov 2026 **The short answer.** LayerX ranks 15th of 16 on the index, and on the Agent Surface Map it has 2 documented cells, 1 partial and 3 none found on 13 Nov 2026. Each cell below carries a note, a source and a read date. Rows follow the site ranking. Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). ## LayerX, cell by cell Desktop apps documented; IDE assistants none found; Command-line agents none found; MCP servers none found; Browser extensions documented; SaaS copilots partial. | Surface | Level | Note | Source and read date | | --- | --- | --- | --- | | Desktop apps | documented | LayerX says a native desktop agent enforces policy on ChatGPT desktop, Claude desktop and Microsoft 365 Copilot desktop (vendor release, May 2026). | [akamai.com](https://www.akamai.com/solutions/ai-usage-control) (12 Nov 2026); [manilatimes.net](https://www.manilatimes.net/2026/05/21/tmt-newswire/globenewswire/layerx-security-integrates-with-new-extended-plan-for-aws-security-hub/2348650) (13 Nov 2026) | | IDE assistants | none found | None found on 12 Nov 2026: product coverage of IDE assistants; only research on Cursor extensions exists. | none found on 12 Nov 2026 | | Command-line agents | none found | None found on 12 Nov 2026: coverage of Claude Code, Codex CLI or Gemini CLI as a product feature. | none found on 12 Nov 2026 | | MCP servers | none found | None found on 12 Nov 2026: local MCP server inventory or control. | none found on 12 Nov 2026 | | Browser extensions | documented | Monitors AI browser agents and scores and blocks risky extensions. | [akamai.com](https://www.akamai.com/products/akamai-workforce-protector) (12 Nov 2026); [akamai.com](https://www.akamai.com/products/akamai-workforce-protector) (12 Nov 2026) | | SaaS copilots | partial | Controls SaaS and web app use through the browser; named SaaS copilots are not listed. | [akamai.com](https://www.akamai.com/products/akamai-workforce-protector) (12 Nov 2026) | LayerX on the Agent Surface Map. LayerX: It enforces warn, redact, block or guide policy on prompts, uploads and copy-paste at the moment of interaction with no proxy or browser replacement, and covers agentic browsers. Rows for every tool: [the Agent Surface Map](/agent-surface-map/). Method: [how a cell is read](/agent-surface-map/methodology/). All [six agent surfaces](/surfaces/), the [full index](/ranking/) and [every head-to-head](/compare/). ## Sources - www.akamai.com/solutions/ai-usage-control: https://www.akamai.com/solutions/ai-usage-control (read 12 Nov 2026) - www.manilatimes.net/2026/05/21/tmt-newswire/globenewswire/layerx-security-integrates-with-new-extended-plan-for-aws-security-hub/2348650: https://www.manilatimes.net/2026/05/21/tmt-newswire/globenewswire/layerx-security-integrates-with-new-extended-plan-for-aws-security-hub/2348650 (read 13 Nov 2026) - www.akamai.com/products/akamai-workforce-protector: https://www.akamai.com/products/akamai-workforce-protector (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/agent-surface-map/layerx/ --- # Palo Alto Networks sees 2 of 6 agent surfaces, with 3 more in part DATASET ROW, read 13 Nov 2026 **The short answer.** Palo Alto Networks ranks 16th of 16 on the index, and on the Agent Surface Map it has 2 documented cells, 3 partial and 1 none found on 13 Nov 2026. Each cell below carries a note, a source and a read date. Rows follow the site ranking. Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). ## Palo Alto Networks, cell by cell Desktop apps partial; IDE assistants documented; Command-line agents partial; MCP servers none found; Browser extensions documented; SaaS copilots partial. | Surface | Level | Note | Source and read date | | --- | --- | --- | --- | | Desktop apps | partial | Prisma AIRS says it discovers AI apps and agents; desktop apps are not named. | [paloaltonetworks.com](https://www.paloaltonetworks.com/prisma/prisma-ai-runtime-security) (12 Nov 2026) | | IDE assistants | documented | AES discovers IDE plugins; AIRS release covers AI coding agents on endpoints. | [paloaltonetworks.com](https://www.paloaltonetworks.com/prisma/prisma-ai-runtime-security) (12 Nov 2026) | | Command-line agents | partial | Coding-agent policy enforcement on endpoints is announced; CLI agents are not named on the page. | [paloaltonetworks.com](https://www.paloaltonetworks.com/prisma/prisma-ai-runtime-security) (12 Nov 2026); [paloaltonetworks.com](https://www.paloaltonetworks.com/blog/2026/10/reinventing-endpoint-security-with-cortex-xdr-one/) (12 Nov 2026) | | MCP servers | none found | None found on 12 Nov 2026 for Prisma AIRS or Prisma Browser; Cortex Agentic Endpoint Security (Koi) coverage is scored on the Koi profile. | none found on 12 Nov 2026 | | Browser extensions | documented | Prisma Browser inventories and blocks risky extensions and sees GenAI and agentic workflows. | [paloaltonetworks.com](https://www.paloaltonetworks.com/sase/prisma-browser) (12 Nov 2026); [paloaltonetworks.com](https://www.paloaltonetworks.com/sase/prisma-browser) (12 Nov 2026) | | SaaS copilots | partial | Prisma Browser controls actions in SaaS and GenAI apps; specific copilots are not named. | [paloaltonetworks.com](https://www.paloaltonetworks.com/sase/prisma-browser) (12 Nov 2026) | Palo Alto Networks on the Agent Surface Map. Palo Alto Networks: Palo Alto Networks documents both a browser-level control point (extension inventory, copy-paste and upload limits for unsanctioned AI apps) and endpoint coding-agent policy enforcement under Prisma AIRS, backed by public docs and analyst recognition. Rows for every tool: [the Agent Surface Map](/agent-surface-map/). Method: [how a cell is read](/agent-surface-map/methodology/). All [six agent surfaces](/surfaces/), the [full index](/ranking/) and [every head-to-head](/compare/). ## Sources - www.paloaltonetworks.com/prisma/prisma-ai-runtime-security: https://www.paloaltonetworks.com/prisma/prisma-ai-runtime-security (read 12 Nov 2026) - www.paloaltonetworks.com/blog/2026/10/reinventing-endpoint-security-with-cortex-xdr-one: https://www.paloaltonetworks.com/blog/2026/10/reinventing-endpoint-security-with-cortex-xdr-one/ (read 12 Nov 2026) - www.paloaltonetworks.com/sase/prisma-browser: https://www.paloaltonetworks.com/sase/prisma-browser (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/agent-surface-map/palo-alto-networks/ --- # Island and Pillar Security can stop the most agent actions: 4 of 6 documented each DATASET, read 13 Nov 2026 **The short answer.** The Action Control Matrix covers 16 tools and 6 actions, 96 cells in all: 31 documented, 42 partial and 23 none found on 13 Nov 2026. Island and Pillar Security have the most documented cells, 4 of 6 each. Rows follow the site ranking. ## Which tools can stop each agent action? - **File read:** stops it: Noma Security, Island, Knostic and Pillar Security; in part: Microsoft, Bay, Aim Security, CrowdStrike and Zenity. - **Shell command:** stops it: Microsoft, Bay, Noma Security, Island, Zenity, Knostic and Pillar Security; in part: Harmonic Security, Aim Security, CrowdStrike, SentinelOne, WitnessAI, Koi, Lasso Security and Palo Alto Networks. - **Network call:** stops it: Bay, Island, Zenity, WitnessAI and Pillar Security; in part: Microsoft, Harmonic Security, Aim Security, CrowdStrike, Noma Security, Knostic, SentinelOne, Koi, Lasso Security and Palo Alto Networks. - **Credential access:** stops it: Bay, Zenity, Knostic and Pillar Security; in part: Microsoft, Harmonic Security, CrowdStrike, Noma Security, Island, SentinelOne, WitnessAI and Lasso Security. - **Data paste:** stops it: Microsoft, Harmonic Security, Aim Security, CrowdStrike, Island, SentinelOne, WitnessAI, Lasso Security, LayerX and Palo Alto Networks; in part: Bay, Noma Security, Zenity, Knostic and Koi. - **Package install:** stops it: CrowdStrike; in part: Bay, Island, Zenity, Knostic, Koi and LayerX. ## The Action Control Matrix as a table Rows follow the site ranking. Cells read documented, partial or none found on 13 Nov 2026. | Tool | Documented | File read | Shell command | Network call | Credential access | Data paste | Package install | | --- | --- | --- | --- | --- | --- | --- | --- | | [Microsoft](/action-control-matrix/microsoft/) | 2 of 6 | partial | documented | partial | partial | documented | none found | | [Bay](/action-control-matrix/bay/) | 3 of 6 | partial | documented | documented | documented | partial | partial | | [Harmonic Security](/action-control-matrix/harmonic-security/) | 1 of 6 | none found | partial | partial | partial | documented | none found | | [Aim Security](/action-control-matrix/aim-security/) | 1 of 6 | partial | partial | partial | none found | documented | none found | | [CrowdStrike](/action-control-matrix/crowdstrike/) | 2 of 6 | partial | partial | partial | partial | documented | documented | | [Noma Security](/action-control-matrix/noma-security/) | 2 of 6 | documented | documented | partial | partial | partial | none found | | [Island](/action-control-matrix/island/) | 4 of 6 | documented | documented | documented | partial | documented | partial | | [Zenity](/action-control-matrix/zenity/) | 3 of 6 | partial | documented | documented | documented | partial | partial | | [Knostic](/action-control-matrix/knostic/) | 3 of 6 | documented | documented | partial | documented | partial | partial | | [SentinelOne](/action-control-matrix/sentinelone/) | 1 of 6 | none found | partial | partial | partial | documented | none found | | [WitnessAI](/action-control-matrix/witnessai/) | 2 of 6 | none found | partial | documented | partial | documented | none found | | [Koi](/action-control-matrix/koi/) | 0 of 6 | none found | partial | partial | none found | partial | partial | | [Lasso Security](/action-control-matrix/lasso-security/) | 1 of 6 | none found | partial | partial | partial | documented | none found | | [Pillar Security](/action-control-matrix/pillar-security/) | 4 of 6 | documented | documented | documented | documented | none found | none found | | [LayerX](/action-control-matrix/layerx/) | 1 of 6 | none found | none found | none found | none found | documented | partial | | [Palo Alto Networks](/action-control-matrix/palo-alto-networks/) | 1 of 6 | none found | partial | partial | none found | documented | none found | Action Control Matrix, edition of 13 Nov 2026. Open a tool's row for each cell's note, source and date. ## Verdict by situation - **Coding agents:** Bay and Microsoft tie at 75.5 and both score 8 on Stops, the heaviest criterion here, so name order puts Bay 1st. Bay has 3 of 6 actions documented; among the top three, Bay documents the most, 3 of 6. - **Regulated data:** Microsoft ranks first at 79.0 with 2 of 6 actions documented; among the top three, Bay documents the most, 3 of 6. Bay ranks 3rd of 16 at 70.5, behind Microsoft at 79.0. Microsoft's lead over Bay comes from policy and governance (9 to 6) and transparency (7 to 1). - **Lean team with EDR:** Bay ranks first at 76.5 with 3 of 6 actions documented; among the top three, Bay documents the most, 3 of 6. - **Staff rollout:** Microsoft ranks first at 75.5 with 2 of 6 actions documented; among the top three, Bay documents the most, 3 of 6. Bay ranks 2nd of 16 at 73.5, behind Microsoft at 75.5. Microsoft's lead over Bay comes from policy and governance (9 to 6) and transparency (7 to 1). - **Board in 30 days:** Bay ranks first at 77.5 with 3 of 6 actions documented; among the top three, Bay documents the most, 3 of 6. Bay is a young company that left stealth in 2026, publishes no price and has little independent testing behind its claims. Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). The Stops score and this matrix measure different things: the matrix counts which of six action types a vendor documents, while the Stops score rates how specific and confirmed the blocking evidence is (action types named, blocking before completion, a second source), so a tool can document fewer action types and still score higher on Stops. ## By action - [File read](/actions/file-read/): 4 of 16 tools documented - [Shell command](/actions/shell-command/): 7 of 16 tools documented - [Network call](/actions/network-call/): 5 of 16 tools documented - [Credential access](/actions/credential-access/): 4 of 16 tools documented - [Data paste](/actions/data-paste/): 10 of 16 tools documented - [Package install](/actions/package-install/): 1 of 16 tools documented ## Download and cite CSV and JSON under CC BY 4.0: [action-control-matrix.csv](/datasets/action-control-matrix.csv) and [action-control-matrix.json](/datasets/action-control-matrix.json). The [methodology](/action-control-matrix/methodology/) explains how a cell is read. Cite this: AI Endpoint Security Ranked, Agent Surface Map, Action Control Matrix and scores, edition of 13 Nov 2026, aiendpointsecurityranked.com/datasets/, CC BY 4.0. ## Sources - Microsoft: learn.microsoft.com/en-us/defender-endpoint/ai-agent-support-matrix: https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-support-matrix (read 12 Nov 2026) - Microsoft: learn.microsoft.com/en-us/microsoft-365/admin/manage/agent-shadow-ai: https://learn.microsoft.com/en-us/microsoft-365/admin/manage/agent-shadow-ai (read 12 Nov 2026) - Microsoft: learn.microsoft.com/en-us/defender-endpoint/ai-agent-runtime-protection-overview: https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-runtime-protection-overview (read 12 Nov 2026) - Microsoft: learn.microsoft.com/en-us/defender-endpoint/local-agent-discovery-overview: https://learn.microsoft.com/en-us/defender-endpoint/local-agent-discovery-overview (read 12 Nov 2026) - Microsoft: learn.microsoft.com/en-us/purview/dspm-for-ai: https://learn.microsoft.com/en-us/purview/dspm-for-ai (read 12 Nov 2026) - Microsoft: learn.microsoft.com/en-us/purview/endpoint-dlp-learn-about: https://learn.microsoft.com/en-us/purview/endpoint-dlp-learn-about (read 12 Nov 2026) - Microsoft: learn.microsoft.com/en-us/purview/ai-microsoft-purview: https://learn.microsoft.com/en-us/purview/ai-microsoft-purview (read 12 Nov 2026) - Microsoft: www.microsoft.com/en-us/security/blog/2026/05/01/microsoft-agent-365-now-generally-available-expands-capabilities-and-integrations: https://www.microsoft.com/en-us/security/blog/2026/05/01/microsoft-agent-365-now-generally-available-expands-capabilities-and-integrations/ (read 12 Nov 2026) - Bay: www.bay.io/platform/ai-endpoint-security: https://www.bay.io/platform/ai-endpoint-security (read 12 Nov 2026) - Bay: softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control: https://softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control (read 12 Nov 2026) - Bay: www.bay.io/platform/cloud-workspaces: https://www.bay.io/platform/cloud-workspaces (read 12 Nov 2026) - Bay: www.bay.io/platform: https://www.bay.io/platform (read 12 Nov 2026) - Bay: www.bay.io/platform/browser: https://www.bay.io/platform/browser (read 12 Nov 2026) - Bay: www.bay.io/blog/bay-emerges-from-stealth-with-50m-after-fortune-100-companies-used-its-platform-to-triple-autonomous-ai-agent-use-2: https://www.bay.io/blog/bay-emerges-from-stealth-with-50m-after-fortune-100-companies-used-its-platform-to-triple-autonomous-ai-agent-use-2 (read 12 Nov 2026) - Harmonic Security: www.harmonic.security/solutions/endpoint-ai-security: https://www.harmonic.security/solutions/endpoint-ai-security (read 12 Nov 2026) - Harmonic Security: docs.harmonicsecurity.app/portal-guides/detection-and-response/sensitive-data-controls-for-coding-agents-and-desktop-applications: https://docs.harmonicsecurity.app/portal-guides/detection-and-response/sensitive-data-controls-for-coding-agents-and-desktop-applications (read 12 Nov 2026) - Harmonic Security: www.harmonic.security/pricing: https://www.harmonic.security/pricing (read 12 Nov 2026) - Harmonic Security: docs.harmonicsecurity.app/portal-guides/managing-your-organization/configure-mcp-server-access-controls: https://docs.harmonicsecurity.app/portal-guides/managing-your-organization/configure-mcp-server-access-controls (read 12 Nov 2026) - Harmonic Security: docs.harmonicsecurity.app/portal-guides/detection-and-response/configure-scenarios-and-interventions: https://docs.harmonicsecurity.app/portal-guides/detection-and-response/configure-scenarios-and-interventions (read 12 Nov 2026) - Harmonic Security: www.harmonic.security/products/command: https://www.harmonic.security/products/command (read 12 Nov 2026) - Harmonic Security: docs.harmonicsecurity.app/llms.txt: https://docs.harmonicsecurity.app/llms.txt (read 12 Nov 2026) - Harmonic Security: docs.harmonicsecurity.app/mcp-gateway/getting-started/what-is-mcp-gateway: https://docs.harmonicsecurity.app/mcp-gateway/getting-started/what-is-mcp-gateway (read 12 Nov 2026) - Harmonic Security: docs.harmonicsecurity.app/portal-guides/integration-guides/configure-anthropic-compliance-api-connector: https://docs.harmonicsecurity.app/portal-guides/integration-guides/configure-anthropic-compliance-api-connector (read 12 Nov 2026) - Aim Security: www.catonetworks.com/platform/cato-ai-endpoint: https://www.catonetworks.com/platform/cato-ai-endpoint/ (read 12 Nov 2026) - Aim Security: www.catonetworks.com/news/cato-acquires-aim-security-to-extend-sase-leadership-and-secure-enterprise-ai-transformation: https://www.catonetworks.com/news/cato-acquires-aim-security-to-extend-sase-leadership-and-secure-enterprise-ai-transformation/ (read 12 Nov 2026) - Aim Security: www.catonetworks.com/platform/agentic-ai-security: https://www.catonetworks.com/platform/agentic-ai-security/ (read 12 Nov 2026) - CrowdStrike: www.msspalert.com/news/crowdstrike-brings-ai-security-to-the-endpoint-and-the-timing-makes-sense: https://www.msspalert.com/news/crowdstrike-brings-ai-security-to-the-endpoint-and-the-timing-makes-sense (read 12 Nov 2026) - CrowdStrike: www.crowdstrike.com/en-us/press-releases/crowdstrike-announces-general-availability-of-falcon-ai-detection-and-response: https://www.crowdstrike.com/en-us/press-releases/crowdstrike-announces-general-availability-of-falcon-ai-detection-and-response/ (read 12 Nov 2026) - CrowdStrike: www.crowdstrike.com/en-us/blog/crowdstrike-advances-ai-security-and-safety-falcon-guardian: https://www.crowdstrike.com/en-us/blog/crowdstrike-advances-ai-security-and-safety-falcon-guardian/ (read 12 Nov 2026) - CrowdStrike: www.crowdstrike.com/en-us/blog/crowdstrike-shadow-AI-visibility-service: https://www.crowdstrike.com/en-us/blog/crowdstrike-shadow-AI-visibility-service/ (read 12 Nov 2026) - CrowdStrike: www.crowdstrike.com/en-us/blog/falcon-guardian-defines-next-generation-of-ai-security: https://www.crowdstrike.com/en-us/blog/falcon-guardian-defines-next-generation-of-ai-security/ (read 12 Nov 2026) - CrowdStrike: www.crowdstrike.com/en-us/platform/falcon-guardian-aidr: https://www.crowdstrike.com/en-us/platform/falcon-guardian-aidr/ (read 12 Nov 2026) - CrowdStrike: www.crowdstrike.com/en-us/blog/crowdstrike-brings-enterprise-browser-security-to-falcon-platform: https://www.crowdstrike.com/en-us/blog/crowdstrike-brings-enterprise-browser-security-to-falcon-platform/ (read 12 Nov 2026) - CrowdStrike: www.crowdstrike.com/en-us/blog/crowdstrike-extends-endpoint-security-to-stop-supply-chain-attacks: https://www.crowdstrike.com/en-us/blog/crowdstrike-extends-endpoint-security-to-stop-supply-chain-attacks/ (read 12 Nov 2026) - Noma Security: noma.security/platform/endpoint-agents: https://noma.security/platform/endpoint-agents (read 12 Nov 2026) - Noma Security: noma.security/blog/noma-brings-agent-security-to-every-employee-endpoint: https://noma.security/blog/noma-brings-agent-security-to-every-employee-endpoint (read 12 Nov 2026) - Noma Security: noma.security/blog/securing-the-agentic-frontier-noma-unveils-the-first-real-time-agent-runtime-security-for-cursor: https://noma.security/blog/securing-the-agentic-frontier-noma-unveils-the-first-real-time-agent-runtime-security-for-cursor (read 12 Nov 2026) - Noma Security: github.com/Noma-Security/noma-marketplace: https://github.com/Noma-Security/noma-marketplace (read 12 Nov 2026) - Noma Security: noma.security/solutions/mcp-security: https://noma.security/solutions/mcp-security (read 12 Nov 2026) - Noma Security: noma.security: https://noma.security/ (read 12 Nov 2026) - Noma Security: github.com/Noma-Security/noma-marketplace/blob/main/guardrails/README.md: https://github.com/Noma-Security/noma-marketplace/blob/main/guardrails/README.md (read 12 Nov 2026) - Island: www.island.io/enterprise-desktop: https://www.island.io/enterprise-desktop (read 12 Nov 2026) - Island: www.island.io/product/ai: https://www.island.io/product/ai (read 12 Nov 2026) - Island: www.island.io/blog/expanding-island-to-govern-your-ai-workforce: https://www.island.io/blog/expanding-island-to-govern-your-ai-workforce (read 12 Nov 2026) - Island: www.island.io/blog/claude-code-cowork-and-claude-for-microsoft-365-now-report-into-island: https://www.island.io/blog/claude-code-cowork-and-claude-for-microsoft-365-now-report-into-island (read 12 Nov 2026) - Island: www.island.io/blog/see-score-and-control-every-ai-asset-on-your-endpoints: https://www.island.io/blog/see-score-and-control-every-ai-asset-on-your-endpoints (read 12 Nov 2026) - Island: www.island.io/product/enterprise-browser-extension: https://www.island.io/product/enterprise-browser-extension (read 12 Nov 2026) - Island: www.island.io/press/island-brings-the-enterprise-browser-to-ai-agents: https://www.island.io/press/island-brings-the-enterprise-browser-to-ai-agents (read 12 Nov 2026) - Island: www.island.io/blog/ai-runtime-protection-governing-what-agents-do-not-just-what-theyre-told: https://www.island.io/blog/ai-runtime-protection-governing-what-agents-do-not-just-what-theyre-told (read 12 Nov 2026) - Island: www.securityweek.com/island-raises-400-million-at-6-4-billion-valuation: https://www.securityweek.com/island-raises-400-million-at-6-4-billion-valuation/ (read 12 Nov 2026) - Island: betanews.com/2025/12/17/governing-ai-where-work-actually-happens-qa: https://betanews.com/2025/12/17/governing-ai-where-work-actually-happens-qa/ (read 12 Nov 2026) - Zenity: zenity.io/use-cases/agent-type/coding-personal-agents: https://zenity.io/use-cases/agent-type/coding-personal-agents (read 12 Nov 2026) - Zenity: zenity.io/blog/coding-agent-attack-surface: https://zenity.io/blog/coding-agent-attack-surface (read 12 Nov 2026) - Zenity: zenity.io/platform/mcp-security: https://zenity.io/platform/mcp-security (read 12 Nov 2026) - Zenity: zenity.io/press-release/zenity-raises-125-million-to-secure-the-era-of-1-billion-ai-agents: https://zenity.io/press-release/zenity-raises-125-million-to-secure-the-era-of-1-billion-ai-agents (read 12 Nov 2026) - Zenity: zenity.io/company: https://zenity.io/company (read 12 Nov 2026) - Zenity: zenity.io: https://zenity.io/ (read 12 Nov 2026) - Zenity: zenity.io/press-release/zenity-extends-ai-agent-security-and-governance-to-claude-enterprise: https://zenity.io/press-release/zenity-extends-ai-agent-security-and-governance-to-claude-enterprise (read 12 Nov 2026) - Zenity: zenity.io/platform/ai-detection-and-response: https://zenity.io/platform/ai-detection-and-response (read 12 Nov 2026) - Knostic: www.knostic.ai/faq: https://www.knostic.ai/faq (read 12 Nov 2026) - Knostic: www.knostic.ai/kirin: https://www.knostic.ai/kirin (read 12 Nov 2026) - Knostic: www.knostic.ai/pricing: https://www.knostic.ai/pricing (read 12 Nov 2026) - Knostic: siliconangle.com/2025/03/05/knostic-raises-11m-strengthen-enterprise-ai-security-need-know-access-controls: https://siliconangle.com/2025/03/05/knostic-raises-11m-strengthen-enterprise-ai-security-need-know-access-controls/ (read 12 Nov 2026) - Knostic: www.knostic.ai: https://www.knostic.ai (read 12 Nov 2026) - SentinelOne: www.sentinelone.com/blog/a-new-chapter-for-ai-and-cybersecurity-sentinelone-acquires-prompt-security: https://www.sentinelone.com/blog/a-new-chapter-for-ai-and-cybersecurity-sentinelone-acquires-prompt-security/ (read 12 Nov 2026) - SentinelOne: www.sentinelone.com/press/sentinelone-expands-ai-security-coverage-to-claude-cowork: https://www.sentinelone.com/press/sentinelone-expands-ai-security-coverage-to-claude-cowork/ (read 12 Nov 2026) - SentinelOne: www.sentinelone.com/blog/prompt-security-for-agentic-ai: https://www.sentinelone.com/blog/prompt-security-for-agentic-ai/ (read 12 Nov 2026) - SentinelOne: www.sentinelone.com/platform/prompt-security-ai-usage-control: https://www.sentinelone.com/platform/prompt-security-ai-usage-control/ (read 13 Nov 2026) - SentinelOne: www.sentinelone.com/press/sentinelone-launches-singularity-agentic-detection-and-response-to-stop-rogue-and-malicious-ai-agents-at-machine-speed: https://www.sentinelone.com/press/sentinelone-launches-singularity-agentic-detection-and-response-to-stop-rogue-and-malicious-ai-agents-at-machine-speed/ (read 12 Nov 2026) - SentinelOne: siliconangle.com/2026/03/23/sentinelone-snyk-introduce-new-tools-securing-ai-agents: https://siliconangle.com/2026/03/23/sentinelone-snyk-introduce-new-tools-securing-ai-agents/ (read 12 Nov 2026) - SentinelOne: prompt.security/blog/the-key-layer-in-ai-security-browser-and-endpoint-sensors: https://prompt.security/blog/the-key-layer-in-ai-security-browser-and-endpoint-sensors (read 12 Nov 2026) - SentinelOne: chromewebstore.google.com/search/prompt%20security: https://chromewebstore.google.com/search/prompt%20security (read 12 Nov 2026) - SentinelOne: www.sentinelone.com/press/sentinelone-named-an-ai-security-platform-leader-in-latio-2026-ai-security-market-report: https://www.sentinelone.com/press/sentinelone-named-an-ai-security-platform-leader-in-latio-2026-ai-security-market-report/ (read 12 Nov 2026) - WitnessAI: witness.ai/for-agents: https://witness.ai/for-agents/ (read 12 Nov 2026) - WitnessAI: witness.ai/product: https://witness.ai/product/ (read 12 Nov 2026) - WitnessAI: witness.ai: https://witness.ai/ (read 12 Nov 2026) - WitnessAI: witness.ai/for-employees: https://witness.ai/for-employees/ (read 12 Nov 2026) - WitnessAI: witness.ai/blog/witnessai-in-sacrs-arise-market-map-why-a-unified-ai-control-plane-matters: https://witness.ai/blog/witnessai-in-sacrs-arise-market-map-why-a-unified-ai-control-plane-matters/ (read 12 Nov 2026) - WitnessAI: witness.ai/protect: https://witness.ai/protect/ (read 12 Nov 2026) - Koi: www.paloaltonetworks.com/resources/techbriefs/koi-agentic-endpoint-security: https://www.paloaltonetworks.com/resources/techbriefs/koi-agentic-endpoint-security (read 12 Nov 2026) - Koi: www.paloaltonetworks.com/resources/datasheets/cortex-agentic-endpoint-security-at-a-glance: https://www.paloaltonetworks.com/resources/datasheets/cortex-agentic-endpoint-security-at-a-glance (read 12 Nov 2026) - Koi: www.paloaltonetworks.com/company/press/2026/palo-alto-networks-completes-acquisition-of-koi-to-secure-the-agentic-endpoint: https://www.paloaltonetworks.com/company/press/2026/palo-alto-networks-completes-acquisition-of-koi-to-secure-the-agentic-endpoint (read 12 Nov 2026) - Koi: www.paloaltonetworks.com/blog/2026/02/securing-the-agentic-endpoint: https://www.paloaltonetworks.com/blog/2026/02/securing-the-agentic-endpoint/ (read 12 Nov 2026) - Koi: www.sdxcentral.com/news/palo-alto-networks-bolsters-ai-security-for-the-endpoint: https://www.sdxcentral.com/news/palo-alto-networks-bolsters-ai-security-for-the-endpoint/ (read 13 Nov 2026) - Koi: cortex-docs.paloaltonetworks.com/cortex-xdr-6.x-rn/readme/features-introduced-in-2026/november-2026/release-highlights.md: https://cortex-docs.paloaltonetworks.com/cortex-xdr-6.x-rn/readme/features-introduced-in-2026/november-2026/release-highlights.md (read 13 Nov 2026) - Koi: www.paloaltonetworks.com/cortex/agentic-endpoint-security.md: https://www.paloaltonetworks.com/cortex/agentic-endpoint-security.md (read 12 Nov 2026) - Koi: www.calcalistech.com/ctechnews/article/xpz78znuc: https://www.calcalistech.com/ctechnews/article/xpz78znuc (read 12 Nov 2026) - Koi: www.securityweek.com/palo-alto-networks-to-acquire-koi-in-reported-400-million-transaction: https://www.securityweek.com/palo-alto-networks-to-acquire-koi-in-reported-400-million-transaction/ (read 12 Nov 2026) - Lasso Security: www.lasso.security/platform/ai-usage-control: https://www.lasso.security/platform/ai-usage-control (read 12 Nov 2026) - Lasso Security: www.lasso.security/blog/securing-ai-on-endpoints-how-to-govern-ai-tools-coding-agents-and-mcp-connections: https://www.lasso.security/blog/securing-ai-on-endpoints-how-to-govern-ai-tools-coding-agents-and-mcp-connections (read 12 Nov 2026) - Lasso Security: www.lasso.security/use-cases/ai-coding-assistants: https://www.lasso.security/use-cases/ai-coding-assistants (read 12 Nov 2026) - Lasso Security: www.lasso.security/use-cases/mcp-security: https://www.lasso.security/use-cases/mcp-security (read 12 Nov 2026) - Lasso Security: github.com/lasso-security/mcp-gateway: https://github.com/lasso-security/mcp-gateway (read 12 Nov 2026) - Lasso Security: www.lasso.security: https://www.lasso.security (read 12 Nov 2026) - Lasso Security: www.lasso.security/blog/bringing-claude-enterprise-activity-into-lasso-with-claudes-compliance-api: https://www.lasso.security/blog/bringing-claude-enterprise-activity-into-lasso-with-claudes-compliance-api (read 12 Nov 2026) - Lasso Security: github.com/lasso-security/kong-ai-lasso-guardrail: https://github.com/lasso-security/kong-ai-lasso-guardrail (read 12 Nov 2026) - Pillar Security: www.pillar.security/platform/agentic-endpoint: https://www.pillar.security/platform/agentic-endpoint (read 12 Nov 2026) - Pillar Security: pulse2.com/vanta-to-acquire-pillar-security-to-expand-ai-agent-protection-as-customer-base-surpasses-20000: https://pulse2.com/vanta-to-acquire-pillar-security-to-expand-ai-agent-protection-as-customer-base-surpasses-20000/ (read 12 Nov 2026) - LayerX: www.akamai.com/solutions/ai-usage-control: https://www.akamai.com/solutions/ai-usage-control (read 12 Nov 2026) - LayerX: www.manilatimes.net/2026/05/21/tmt-newswire/globenewswire/layerx-security-integrates-with-new-extended-plan-for-aws-security-hub/2348650: https://www.manilatimes.net/2026/05/21/tmt-newswire/globenewswire/layerx-security-integrates-with-new-extended-plan-for-aws-security-hub/2348650 (read 13 Nov 2026) - LayerX: www.akamai.com/products/akamai-workforce-protector: https://www.akamai.com/products/akamai-workforce-protector (read 12 Nov 2026) - Palo Alto Networks: www.paloaltonetworks.com/prisma/prisma-ai-runtime-security: https://www.paloaltonetworks.com/prisma/prisma-ai-runtime-security (read 12 Nov 2026) - Palo Alto Networks: www.paloaltonetworks.com/blog/2026/10/reinventing-endpoint-security-with-cortex-xdr-one: https://www.paloaltonetworks.com/blog/2026/10/reinventing-endpoint-security-with-cortex-xdr-one/ (read 12 Nov 2026) - Palo Alto Networks: www.paloaltonetworks.com/sase/prisma-browser: https://www.paloaltonetworks.com/sase/prisma-browser (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/action-control-matrix/ --- # A Action Control Matrix cell is yes only when public documentation names the action METHODOLOGY, read 13 Nov 2026 **The short answer.** Each cell of the Action Control Matrix answers which agent actions it can stop before they complete, as documented on 13 Nov 2026. Yes needs a documented capability that names the action, partial covers a limited or conditional case, and none found means no public source named it. The matrix holds 96 cells across 16 tools. ## What the levels mean - **Yes:** The vendor documents blocking or denying this action before it completes, naming it. - **Partial:** The documentation names a limited case, a condition, a roadmap item or detection after the fact. - **None found:** No public source read on 13 Nov 2026 names it. This is written "none found on 13 Nov 2026" and is not a claim about the product. ## The 6 actions - **File read:** An agent opening a local file, including files outside the project it was started in. - **Shell command:** An agent running a command in a shell on the endpoint. - **Network call:** An agent sending a request to an outside host or service. - **Credential access:** An agent reading keys, tokens, cookies or password stores. - **Data paste:** A user or agent pasting or uploading sensitive text into an AI tool. - **Package install:** An agent installing a package, plugin, skill or extension. ## How a cell is filled - Each cell cites the fact ids that support it in the vendor evidence file, with a quote, a source URL and a read date. - A vendor's own page counts as the vendor's claim and is labelled that way; a cell stays yes or partial on a vendor claim alone, while the criterion scores cap such claims at 7 of 10. - Cells were read on 13 Nov 2026. A cell changes only when a sourced fact changes, and the change is logged with the fact. - Rows are ordered by the site ranking, which the cells inform but do not decide: the Sees and Stops criteria also weigh depth and second sources. Across the 16 tools there are 31 documented cells, 42 partial cells and 23 cells where nothing was found on 13 Nov 2026. Cite this: AI Endpoint Security Ranked, Agent Surface Map, Action Control Matrix and scores, edition of 13 Nov 2026, aiendpointsecurityranked.com/datasets/, CC BY 4.0. ## Use the cells - [Open the Action Control Matrix](/action-control-matrix/) - [The six agent actions](/actions/) - [Download the datasets](/datasets/) - [The scoring method](/method/) ## Sources - Action Control Matrix: https://aiendpointsecurityranked.com/action-control-matrix/ (read 13 Nov 2026) - Scoring method: https://aiendpointsecurityranked.com/method/ (read 13 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/action-control-matrix/methodology/ --- # Microsoft can stop 2 of 6 agent actions, with 3 more in part DATASET ROW, read 13 Nov 2026 **The short answer.** Microsoft ranks 1st of 16 on the index, and on the Action Control Matrix it has 2 documented cells, 3 partial and 1 none found on 13 Nov 2026. Each cell below carries a note, a source and a read date. Rows follow the site ranking. Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). ## Microsoft, cell by cell File read partial; Shell command documented; Network call partial; Credential access partial; Data paste documented; Package install none found. | Action | Level | Note | Source and read date | | --- | --- | --- | --- | | File read | partial | Example blocks reading a local environment file after injection; general not file-specific. | [learn.microsoft.com](https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-runtime-protection-overview) (12 Nov 2026); [learn.microsoft.com](https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-runtime-protection-overview) (12 Nov 2026) | | Shell command | documented | Agent-native inspection can prevent a harmful tool action from running in supported agents (preview, Windows). | [learn.microsoft.com](https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-runtime-protection-overview) (12 Nov 2026); [learn.microsoft.com](https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-runtime-protection-overview) (12 Nov 2026) | | Network call | partial | Blocks data leaving after injection; network inspection excludes pinned and HTTP/3 agents. | [learn.microsoft.com](https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-runtime-protection-overview) (12 Nov 2026); [learn.microsoft.com](https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-runtime-protection-overview) (12 Nov 2026) | | Credential access | partial | Example covers environment file contents; credential-specific blocking not documented. | [learn.microsoft.com](https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-runtime-protection-overview) (12 Nov 2026); [learn.microsoft.com](https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-runtime-protection-overview) (12 Nov 2026) | | Data paste | documented | Endpoint DLP detects paste to restricted domains and blocks uploads from unallowed browsers. | [learn.microsoft.com](https://learn.microsoft.com/en-us/purview/endpoint-dlp-learn-about) (12 Nov 2026); [learn.microsoft.com](https://learn.microsoft.com/en-us/purview/endpoint-dlp-learn-about) (12 Nov 2026) | | Package install | none found | None found on 12 Nov 2026 in Defender AI agent docs. | none found on 12 Nov 2026 | Microsoft on the Action Control Matrix. Microsoft: Defender discovers dozens of local agents and MCP servers on Windows and macOS (GA) and can block prompt-injection-driven actions in Claude Code, Codex CLI and GitHub Copilot CLI on Windows, with Endpoint DLP blocking pastes and uploads to AI sites. Rows for every tool: [the Action Control Matrix](/action-control-matrix/). Method: [how a cell is read](/action-control-matrix/methodology/). All [six agent actions](/actions/), the [full index](/ranking/) and [every head-to-head](/compare/). ## Sources - learn.microsoft.com/en-us/defender-endpoint/ai-agent-runtime-protection-overview: https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-runtime-protection-overview (read 12 Nov 2026) - learn.microsoft.com/en-us/purview/endpoint-dlp-learn-about: https://learn.microsoft.com/en-us/purview/endpoint-dlp-learn-about (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/action-control-matrix/microsoft/ --- # Bay can stop 3 of 6 agent actions, with 3 more in part DATASET ROW, read 13 Nov 2026 **The short answer.** Bay ranks 2nd of 16 on the index, and on the Action Control Matrix it has 3 documented cells, 3 partial and 0 none found on 13 Nov 2026. Each cell below carries a note, a source and a read date. Rows follow the site ranking. ## Bay, cell by cell File read partial; Shell command documented; Network call documented; Credential access documented; Data paste partial; Package install partial. | Action | Level | Note | Source and read date | | --- | --- | --- | --- | | File read | partial | Analyst profile lists files among evaluated action types; Bay pages do not detail file-read blocking. | [softwareanalyst.substack.com](https://softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control) (12 Nov 2026) | | Shell command | documented | Bay names risky shell commands as an action it allows, asks, denies or steers. | [bay.io](https://www.bay.io/platform/ai-endpoint-security?utm_source=aiendpointsecurityranked.com&utm_medium=referral&utm_campaign=asset) (12 Nov 2026) | | Network call | documented | Analyst profile lists network destinations and exfiltration-route closure, as stated by Bay. | [softwareanalyst.substack.com](https://softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control) (12 Nov 2026); [softwareanalyst.substack.com](https://softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control) (12 Nov 2026) | | Credential access | documented | Bay names credential access as an action it controls. | [bay.io](https://www.bay.io/platform/ai-endpoint-security?utm_source=aiendpointsecurityranked.com&utm_medium=referral&utm_campaign=asset) (12 Nov 2026); [softwareanalyst.substack.com](https://softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control) (12 Nov 2026) | | Data paste | partial | Bay says browser guardrails keep personal data, secrets and source code out of prompts; detail on paste or upload blocking not found. | [bay.io](https://www.bay.io/platform/browser?utm_source=aiendpointsecurityranked.com&utm_medium=referral&utm_campaign=asset) (12 Nov 2026) | | Package install | partial | Analyst profile reports approve, block and uninstall controls for packages and extensions; vendor-reported. | [softwareanalyst.substack.com](https://softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control) (12 Nov 2026); [softwareanalyst.substack.com](https://softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control) (12 Nov 2026) | Bay on the Action Control Matrix. Bay: Bay documents on-device, session-aware allow/ask/deny/steer enforcement with named action types, agentless MDM/EDR deployment in under 5 minutes (Bay's claim), and one policy across endpoint, browser and cloud. Rows for every tool: [the Action Control Matrix](/action-control-matrix/). Method: [how a cell is read](/action-control-matrix/methodology/). **About Bay.** Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). All [six agent actions](/actions/), the [full index](/ranking/) and [every head-to-head](/compare/). ## Sources - softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control: https://softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control (read 12 Nov 2026) - www.bay.io/platform/ai-endpoint-security: https://www.bay.io/platform/ai-endpoint-security (read 12 Nov 2026) - www.bay.io/platform/browser: https://www.bay.io/platform/browser (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/action-control-matrix/bay/ --- # Harmonic Security can stop 1 of 6 agent actions, with 3 more in part DATASET ROW, read 13 Nov 2026 **The short answer.** Harmonic Security ranks 3rd of 16 on the index, and on the Action Control Matrix it has 1 documented cells, 3 partial and 2 none found on 13 Nov 2026. Each cell below carries a note, a source and a read date. Rows follow the site ranking. Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). ## Harmonic Security, cell by cell File read none found; Shell command partial; Network call partial; Credential access partial; Data paste documented; Package install none found. | Action | Level | Note | Source and read date | | --- | --- | --- | --- | | File read | none found | Documented controls cover prompts and tool calls; file reads not named. | [docs.harmonicsecurity.app](https://docs.harmonicsecurity.app/portal-guides/detection-and-response/sensitive-data-controls-for-coding-agents-and-desktop-applications) (12 Nov 2026) | | Shell command | partial | Tool calls are intercepted through hooks, shell commands not named specifically. | [docs.harmonicsecurity.app](https://docs.harmonicsecurity.app/portal-guides/detection-and-response/sensitive-data-controls-for-coding-agents-and-desktop-applications) (12 Nov 2026) | | Network call | partial | MCP gateway blocks unpermitted server access and tool calls; general network calls not named. | [docs.harmonicsecurity.app](https://docs.harmonicsecurity.app/portal-guides/managing-your-organization/configure-mcp-server-access-controls) (12 Nov 2026); [harmonic.security](https://www.harmonic.security/pricing) (12 Nov 2026) | | Credential access | partial | Sensitive data detection blocks credentials in prompts; credential-file access not named. | [harmonic.security](https://www.harmonic.security/pricing) (12 Nov 2026) | | Data paste | documented | Prompts with sensitive data can be blocked in browser, coding agents and desktop apps. | [docs.harmonicsecurity.app](https://docs.harmonicsecurity.app/portal-guides/detection-and-response/sensitive-data-controls-for-coding-agents-and-desktop-applications) (12 Nov 2026); [harmonic.security](https://www.harmonic.security/pricing) (12 Nov 2026) | | Package install | none found | No package install controls found. | [docs.harmonicsecurity.app](https://docs.harmonicsecurity.app/portal-guides/detection-and-response/sensitive-data-controls-for-coding-agents-and-desktop-applications) (12 Nov 2026) | Harmonic Security on the Action Control Matrix. Harmonic Security: Harmonic documents hook-based inline blocking of prompts and tool calls across named coding agents (Claude Code, Codex, Cursor, Copilot CLI and others) plus an MCP gateway with identity-aware access groups. Rows for every tool: [the Action Control Matrix](/action-control-matrix/). Method: [how a cell is read](/action-control-matrix/methodology/). All [six agent actions](/actions/), the [full index](/ranking/) and [every head-to-head](/compare/). ## Sources - docs.harmonicsecurity.app/portal-guides/detection-and-response/sensitive-data-controls-for-coding-agents-and-desktop-applications (checked, nothing found): https://docs.harmonicsecurity.app/portal-guides/detection-and-response/sensitive-data-controls-for-coding-agents-and-desktop-applications (read 12 Nov 2026) - docs.harmonicsecurity.app/portal-guides/managing-your-organization/configure-mcp-server-access-controls: https://docs.harmonicsecurity.app/portal-guides/managing-your-organization/configure-mcp-server-access-controls (read 12 Nov 2026) - www.harmonic.security/pricing: https://www.harmonic.security/pricing (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/action-control-matrix/harmonic-security/ --- # Aim Security can stop 1 of 6 agent actions, with 3 more in part DATASET ROW, read 13 Nov 2026 **The short answer.** Aim Security ranks 4th of 16 on the index, and on the Action Control Matrix it has 1 documented cells, 3 partial and 2 none found on 13 Nov 2026. Each cell below carries a note, a source and a read date. Rows follow the site ranking. Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). ## Aim Security, cell by cell File read partial; Shell command partial; Network call partial; Credential access none found; Data paste documented; Package install none found. | Action | Level | Note | Source and read date | | --- | --- | --- | --- | | File read | partial | Dashboard flags risky settings such as writes outside the project; no documented block of file reads. | [catonetworks.com](https://www.catonetworks.com/platform/cato-ai-endpoint/) (12 Nov 2026) | | Shell command | partial | Flags full command-line access; blocks unapproved tool calls in supported agents, not shell commands specifically. | [catonetworks.com](https://www.catonetworks.com/platform/cato-ai-endpoint/) (12 Nov 2026); [catonetworks.com](https://www.catonetworks.com/platform/agentic-ai-security/) (12 Nov 2026) | | Network call | partial | Inline network inspection of AI traffic and tool-call blocking; no documented blocking of arbitrary agent network calls. | [catonetworks.com](https://www.catonetworks.com/platform/cato-ai-endpoint/) (12 Nov 2026); [catonetworks.com](https://www.catonetworks.com/platform/agentic-ai-security/) (12 Nov 2026) | | Credential access | none found | None found on 12 Nov 2026. | none found on 12 Nov 2026 | | Data paste | documented | Blocks sensitive prompts at send for desktop AI and can monitor, anonymize or block in the browser. | [catonetworks.com](https://www.catonetworks.com/platform/cato-ai-endpoint/) (12 Nov 2026); [catonetworks.com](https://www.catonetworks.com/platform/cato-ai-endpoint/) (12 Nov 2026) | | Package install | none found | None found on 12 Nov 2026. | none found on 12 Nov 2026 | Aim Security on the Action Control Matrix. Aim Security: Cato AI Endpoint documents a lightweight client that finds local agents with their MCP servers and skills and blocks an unapproved tool call before it runs, in one dashboard with inline network inspection. Rows for every tool: [the Action Control Matrix](/action-control-matrix/). Method: [how a cell is read](/action-control-matrix/methodology/). All [six agent actions](/actions/), the [full index](/ranking/) and [every head-to-head](/compare/). ## Sources - www.catonetworks.com/platform/cato-ai-endpoint: https://www.catonetworks.com/platform/cato-ai-endpoint/ (read 12 Nov 2026) - www.catonetworks.com/platform/agentic-ai-security: https://www.catonetworks.com/platform/agentic-ai-security/ (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/action-control-matrix/aim-security/ --- # CrowdStrike can stop 2 of 6 agent actions, with 4 more in part DATASET ROW, read 13 Nov 2026 **The short answer.** CrowdStrike ranks 5th of 16 on the index, and on the Action Control Matrix it has 2 documented cells, 4 partial and 0 none found on 13 Nov 2026. Each cell below carries a note, a source and a read date. Rows follow the site ranking. Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). ## CrowdStrike, cell by cell File read partial; Shell command partial; Network call partial; Credential access partial; Data paste documented; Package install documented. | Action | Level | Note | Source and read date | | --- | --- | --- | --- | | File read | partial | Sensor sees files agents access; blocking is documented via agent hooks in general terms. | [crowdstrike.com](https://www.crowdstrike.com/en-us/platform/falcon-guardian-aidr/) (12 Nov 2026); [crowdstrike.com](https://www.crowdstrike.com/en-us/blog/crowdstrike-advances-ai-security-and-safety-falcon-guardian/) (12 Nov 2026) | | Shell command | partial | Sensor sees commands agents run; hooks can block malicious activity before execution in supported agents. | [crowdstrike.com](https://www.crowdstrike.com/en-us/platform/falcon-guardian-aidr/) (12 Nov 2026); [crowdstrike.com](https://www.crowdstrike.com/en-us/blog/crowdstrike-advances-ai-security-and-safety-falcon-guardian/) (12 Nov 2026) | | Network call | partial | Kill switch blocks model access at a gateway; endpoint network inspection is pre-beta. | [crowdstrike.com](https://www.crowdstrike.com/en-us/platform/falcon-guardian-aidr/) (12 Nov 2026); [crowdstrike.com](https://www.crowdstrike.com/en-us/blog/crowdstrike-advances-ai-security-and-safety-falcon-guardian/) (12 Nov 2026) | | Credential access | partial | Credentials in prompts are blocked before reaching models; credential file access blocking not specifically documented. | [crowdstrike.com](https://www.crowdstrike.com/en-us/press-releases/crowdstrike-announces-general-availability-of-falcon-ai-detection-and-response/) (12 Nov 2026); [crowdstrike.com](https://www.crowdstrike.com/en-us/blog/crowdstrike-advances-ai-security-and-safety-falcon-guardian/) (12 Nov 2026) | | Data paste | documented | Sensitive data blocked before reaching models; Seraphic enforces paste and upload policy in the browser. | [crowdstrike.com](https://www.crowdstrike.com/en-us/press-releases/crowdstrike-announces-general-availability-of-falcon-ai-detection-and-response/) (12 Nov 2026); [crowdstrike.com](https://www.crowdstrike.com/en-us/blog/crowdstrike-brings-enterprise-browser-security-to-falcon-platform/) (12 Nov 2026) | | Package install | documented | Sensor blocks malicious npm and PyPI packages before embedded code can execute. | [crowdstrike.com](https://www.crowdstrike.com/en-us/blog/crowdstrike-extends-endpoint-security-to-stop-supply-chain-attacks/) (12 Nov 2026); [crowdstrike.com](https://www.crowdstrike.com/en-us/blog/crowdstrike-extends-endpoint-security-to-stop-supply-chain-attacks/) (12 Nov 2026) | CrowdStrike on the Action Control Matrix. CrowdStrike: Falcon Guardian pairs OS-level sensor telemetry with agent-level hooks that can block actions before they run in Claude Code, Codex, GitHub Copilot, OpenCode and Cursor, plus package blocking in the same sensor. Rows for every tool: [the Action Control Matrix](/action-control-matrix/). Method: [how a cell is read](/action-control-matrix/methodology/). All [six agent actions](/actions/), the [full index](/ranking/) and [every head-to-head](/compare/). ## Sources - www.crowdstrike.com/en-us/platform/falcon-guardian-aidr: https://www.crowdstrike.com/en-us/platform/falcon-guardian-aidr/ (read 12 Nov 2026) - www.crowdstrike.com/en-us/blog/crowdstrike-advances-ai-security-and-safety-falcon-guardian: https://www.crowdstrike.com/en-us/blog/crowdstrike-advances-ai-security-and-safety-falcon-guardian/ (read 12 Nov 2026) - www.crowdstrike.com/en-us/press-releases/crowdstrike-announces-general-availability-of-falcon-ai-detection-and-response: https://www.crowdstrike.com/en-us/press-releases/crowdstrike-announces-general-availability-of-falcon-ai-detection-and-response/ (read 12 Nov 2026) - www.crowdstrike.com/en-us/blog/crowdstrike-brings-enterprise-browser-security-to-falcon-platform: https://www.crowdstrike.com/en-us/blog/crowdstrike-brings-enterprise-browser-security-to-falcon-platform/ (read 12 Nov 2026) - www.crowdstrike.com/en-us/blog/crowdstrike-extends-endpoint-security-to-stop-supply-chain-attacks: https://www.crowdstrike.com/en-us/blog/crowdstrike-extends-endpoint-security-to-stop-supply-chain-attacks/ (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/action-control-matrix/crowdstrike/ --- # Noma Security can stop 2 of 6 agent actions, with 3 more in part DATASET ROW, read 13 Nov 2026 **The short answer.** Noma Security ranks 6th of 16 on the index, and on the Action Control Matrix it has 2 documented cells, 3 partial and 1 none found on 13 Nov 2026. Each cell below carries a note, a source and a read date. Rows follow the site ranking. Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). ## Noma Security, cell by cell File read documented; Shell command documented; Network call partial; Credential access partial; Data paste partial; Package install none found. | Action | Level | Note | Source and read date | | --- | --- | --- | --- | | File read | documented | Noma's public Claude Code plugin README documents hook-based evaluation of file reads (.env, SSH keys) with block, mask and ask policies (beta; fail-open on hook failure). | [noma.security](https://noma.security/blog/noma-brings-agent-security-to-every-employee-endpoint) (12 Nov 2026); [github.com](https://github.com/Noma-Security/noma-marketplace/blob/main/guardrails/README.md) (12 Nov 2026) | | Shell command | documented | Noma's public Claude Code plugin README documents hook-based evaluation of shell commands with block, mask and ask policies (beta; fail-open on hook failure). | [noma.security](https://noma.security/blog/securing-the-agentic-frontier-noma-unveils-the-first-real-time-agent-runtime-security-for-cursor) (12 Nov 2026); [noma.security](https://noma.security/platform/endpoint-agents) (12 Nov 2026) | | Network call | partial | Block and route-to-human decisions at the action level; network-specific control not itemized. | [noma.security](https://noma.security/blog/noma-brings-agent-security-to-every-employee-endpoint) (12 Nov 2026); [noma.security](https://noma.security/platform/endpoint-agents) (12 Nov 2026) | | Credential access | partial | Credential leakage named as a risk and secrets detection in AI-DR; block per detector setting. | [noma.security](https://noma.security/blog/noma-brings-agent-security-to-every-employee-endpoint) (12 Nov 2026); [noma.security](https://noma.security/blog/securing-the-agentic-frontier-noma-unveils-the-first-real-time-agent-runtime-security-for-cursor) (12 Nov 2026) | | Data paste | partial | Inline masking of sensitive data is documented for agent traffic; no paste-into-AI-tool control found. | [noma.security](https://noma.security/blog/noma-brings-agent-security-to-every-employee-endpoint) (12 Nov 2026) | | Package install | none found | Nothing found on 12 Nov 2026 on blocking package installs by agents. | none found on 12 Nov 2026 | Noma Security on the Action Control Matrix. Noma Security: Has a dedicated, current endpoint page: discovery of agents, MCP servers and skills via existing EDR or MDM, plus action-level allow and block policy tied to IdP identity. Rows for every tool: [the Action Control Matrix](/action-control-matrix/). Method: [how a cell is read](/action-control-matrix/methodology/). All [six agent actions](/actions/), the [full index](/ranking/) and [every head-to-head](/compare/). ## Sources - noma.security/blog/noma-brings-agent-security-to-every-employee-endpoint: https://noma.security/blog/noma-brings-agent-security-to-every-employee-endpoint (read 12 Nov 2026) - github.com/Noma-Security/noma-marketplace/blob/main/guardrails/README.md: https://github.com/Noma-Security/noma-marketplace/blob/main/guardrails/README.md (read 12 Nov 2026) - noma.security/blog/securing-the-agentic-frontier-noma-unveils-the-first-real-time-agent-runtime-security-for-cursor: https://noma.security/blog/securing-the-agentic-frontier-noma-unveils-the-first-real-time-agent-runtime-security-for-cursor (read 12 Nov 2026) - noma.security/platform/endpoint-agents: https://noma.security/platform/endpoint-agents (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/action-control-matrix/noma-security/ --- # Island can stop 4 of 6 agent actions, with 2 more in part DATASET ROW, read 13 Nov 2026 **The short answer.** Island ranks 7th of 16 on the index, and on the Action Control Matrix it has 4 documented cells, 2 partial and 0 none found on 13 Nov 2026. Each cell below carries a note, a source and a read date. Rows follow the site ranking. Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). ## Island, cell by cell File read documented; Shell command documented; Network call documented; Credential access partial; Data paste documented; Package install partial. | Action | Level | Note | Source and read date | | --- | --- | --- | --- | | File read | documented | Says the file read on the endpoint can be blocked in an agent chain. | [island.io](https://www.island.io/blog/claude-code-cowork-and-claude-for-microsoft-365-now-report-into-island) (12 Nov 2026) | | Shell command | documented | Bash execution can be blocked, or the user can be prompted, per policy. | [island.io](https://www.island.io/product/ai) (12 Nov 2026); [island.io](https://www.island.io/blog/ai-runtime-protection-governing-what-agents-do-not-just-what-theyre-told) (12 Nov 2026) | | Network call | documented | Says the outbound connection can be blocked on the network; malicious URLs are also blocked. | [island.io](https://www.island.io/blog/claude-code-cowork-and-claude-for-microsoft-365-now-report-into-island) (12 Nov 2026) | | Credential access | partial | Time-boxed credentials and a browser vault are described; blocking of local credential file access is not documented. | [island.io](https://www.island.io/product/ai) (12 Nov 2026); [island.io](https://www.island.io/press/island-brings-the-enterprise-browser-to-ai-agents) (12 Nov 2026) | | Data paste | documented | Clipboard, file and USB DLP on the endpoint; prompt and upload DLP in the browser. | [island.io](https://www.island.io/enterprise-desktop) (12 Nov 2026); [island.io](https://www.island.io/blog/ai-runtime-protection-governing-what-agents-do-not-just-what-theyre-told) (12 Nov 2026) | | Package install | partial | Code packages are scanned and flagged installs can be removed; pre-install blocking is not specified. | [island.io](https://www.island.io/product/ai) (12 Nov 2026); [island.io](https://www.island.io/blog/expanding-island-to-govern-your-ai-workforce) (12 Nov 2026) | Island on the Action Control Matrix. Island: Island combines an endpoint agent for Windows and macOS with a browser, extension and MCP gateway, and documents blocking of Bash execution, file reads and network connections as well as clipboard DLP. Rows for every tool: [the Action Control Matrix](/action-control-matrix/). Method: [how a cell is read](/action-control-matrix/methodology/). All [six agent actions](/actions/), the [full index](/ranking/) and [every head-to-head](/compare/). ## Sources - www.island.io/blog/claude-code-cowork-and-claude-for-microsoft-365-now-report-into-island: https://www.island.io/blog/claude-code-cowork-and-claude-for-microsoft-365-now-report-into-island (read 12 Nov 2026) - www.island.io/product/ai: https://www.island.io/product/ai (read 12 Nov 2026) - www.island.io/blog/ai-runtime-protection-governing-what-agents-do-not-just-what-theyre-told: https://www.island.io/blog/ai-runtime-protection-governing-what-agents-do-not-just-what-theyre-told (read 12 Nov 2026) - www.island.io/press/island-brings-the-enterprise-browser-to-ai-agents: https://www.island.io/press/island-brings-the-enterprise-browser-to-ai-agents (read 12 Nov 2026) - www.island.io/enterprise-desktop: https://www.island.io/enterprise-desktop (read 12 Nov 2026) - www.securityweek.com/island-raises-400-million-at-6-4-billion-valuation: https://www.securityweek.com/island-raises-400-million-at-6-4-billion-valuation/ (read 12 Nov 2026) - betanews.com/2025/12/17/governing-ai-where-work-actually-happens-qa: https://betanews.com/2025/12/17/governing-ai-where-work-actually-happens-qa/ (read 12 Nov 2026) - www.island.io/blog/expanding-island-to-govern-your-ai-workforce: https://www.island.io/blog/expanding-island-to-govern-your-ai-workforce (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/action-control-matrix/island/ --- # Zenity can stop 3 of 6 agent actions, with 3 more in part DATASET ROW, read 13 Nov 2026 **The short answer.** Zenity ranks 8th of 16 on the index, and on the Action Control Matrix it has 3 documented cells, 3 partial and 0 none found on 13 Nov 2026. Each cell below carries a note, a source and a read date. Rows follow the site ranking. Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). ## Zenity, cell by cell File read partial; Shell command documented; Network call documented; Credential access documented; Data paste partial; Package install partial. | Action | Level | Note | Source and read date | | --- | --- | --- | --- | | File read | partial | Sensitive-file access followed by network action is blocked; plain file reads are monitored. | [zenity.io](https://zenity.io/blog/coding-agent-attack-surface) (12 Nov 2026) | | Shell command | documented | Block or modify a dangerous tool call (commands) before execution via hooks. | [zenity.io](https://zenity.io/use-cases/agent-type/coding-personal-agents) (12 Nov 2026); [zenity.io](https://zenity.io/platform/ai-detection-and-response) (12 Nov 2026) | | Network call | documented | Network-bound action after sensitive access is blocked at PreToolUse. | [zenity.io](https://zenity.io/blog/coding-agent-attack-surface) (12 Nov 2026) | | Credential access | documented | Secret exfiltration prevention named.env, SSH keys, credential stores in the blog. | [zenity.io](https://zenity.io/blog/coding-agent-attack-surface) (12 Nov 2026); [zenity.io](https://zenity.io/use-cases/agent-type/coding-personal-agents) (12 Nov 2026) | | Data paste | partial | Governance controls on data destinations; no paste or upload-to-chat control documented for desktop AI apps. | [zenity.io](https://zenity.io/use-cases/agent-type/coding-personal-agents) (12 Nov 2026); [zenity.io](https://zenity.io/platform/ai-detection-and-response) (12 Nov 2026) | | Package install | partial | Malicious package install prevention named on the product page; details sit in JS-only docs. | [zenity.io](https://zenity.io/use-cases/agent-type/coding-personal-agents) (12 Nov 2026) | Zenity on the Action Control Matrix. Zenity: Documents pre-execution blocking of coding-agent tool calls through native hooks and an MCP gateway, including credential-read-then-network sequences. Rows for every tool: [the Action Control Matrix](/action-control-matrix/). Method: [how a cell is read](/action-control-matrix/methodology/). All [six agent actions](/actions/), the [full index](/ranking/) and [every head-to-head](/compare/). ## Sources - zenity.io/blog/coding-agent-attack-surface: https://zenity.io/blog/coding-agent-attack-surface (read 12 Nov 2026) - zenity.io/use-cases/agent-type/coding-personal-agents: https://zenity.io/use-cases/agent-type/coding-personal-agents (read 12 Nov 2026) - zenity.io/platform/ai-detection-and-response: https://zenity.io/platform/ai-detection-and-response (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/action-control-matrix/zenity/ --- # Knostic can stop 3 of 6 agent actions, with 3 more in part DATASET ROW, read 13 Nov 2026 **The short answer.** Knostic ranks 9th of 16 on the index, and on the Action Control Matrix it has 3 documented cells, 3 partial and 0 none found on 13 Nov 2026. Each cell below carries a note, a source and a read date. Rows follow the site ranking. Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). ## Knostic, cell by cell File read documented; Shell command documented; Network call partial; Credential access documented; Data paste partial; Package install partial. | Action | Level | Note | Source and read date | | --- | --- | --- | --- | | File read | documented | Enforce mode blocks reads of files containing secrets. | [knostic.ai](https://www.knostic.ai/faq) (12 Nov 2026) | | Shell command | documented | Demo shows destructive command blocked before it ran; Kirin page claims block of risky commands. | [knostic.ai](https://www.knostic.ai) (12 Nov 2026); [knostic.ai](https://www.knostic.ai/kirin) (12 Nov 2026) | | Network call | partial | Risky MCP servers quarantined so agents cannot connect. | [knostic.ai](https://www.knostic.ai/faq) (12 Nov 2026) | | Credential access | documented | Secret reads blocked or sanitized; secrets in prompts blocked by hook. | [knostic.ai](https://www.knostic.ai/faq) (12 Nov 2026); [knostic.ai](https://www.knostic.ai/kirin) (12 Nov 2026) | | Data paste | partial | Secrets in prompts inside agents blocked; browser chatbot use is visibility-only. | [knostic.ai](https://www.knostic.ai/kirin) (12 Nov 2026) | | Package install | partial | Dependency inspection and malicious package alerts; blocking in Enforce is claimed generally. | [knostic.ai](https://www.knostic.ai/kirin) (12 Nov 2026) | Knostic on the Action Control Matrix. Knostic: Knostic documents Enforce-mode blocking of secret file reads, risky MCP servers, destructive commands and unsafe packages inside agents, with a published $15 per user Team price and free tier. Rows for every tool: [the Action Control Matrix](/action-control-matrix/). Method: [how a cell is read](/action-control-matrix/methodology/). All [six agent actions](/actions/), the [full index](/ranking/) and [every head-to-head](/compare/). ## Sources - www.knostic.ai/faq: https://www.knostic.ai/faq (read 12 Nov 2026) - www.knostic.ai: https://www.knostic.ai (read 12 Nov 2026) - www.knostic.ai/kirin: https://www.knostic.ai/kirin (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/action-control-matrix/knostic/ --- # SentinelOne can stop 1 of 6 agent actions, with 3 more in part DATASET ROW, read 13 Nov 2026 **The short answer.** SentinelOne ranks 10th of 16 on the index, and on the Action Control Matrix it has 1 documented cells, 3 partial and 2 none found on 13 Nov 2026. Each cell below carries a note, a source and a read date. Rows follow the site ranking. Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). ## SentinelOne, cell by cell File read none found; Shell command partial; Network call partial; Credential access partial; Data paste documented; Package install none found. | Action | Level | Note | Source and read date | | --- | --- | --- | --- | | File read | none found | None found on 12 Nov 2026. | none found on 12 Nov 2026 | | Shell command | partial | Tool-call blocking and a Cowork VM behavioral engine; shell commands not named. | [sentinelone.com](https://www.sentinelone.com/blog/prompt-security-for-agentic-ai/) (12 Nov 2026); [sentinelone.com](https://www.sentinelone.com/press/sentinelone-expands-ai-security-coverage-to-claude-cowork/) (12 Nov 2026) | | Network call | partial | MCP calls are intercepted and can be blocked; other network calls not named. | [sentinelone.com](https://www.sentinelone.com/blog/a-new-chapter-for-ai-and-cybersecurity-sentinelone-acquires-prompt-security/) (12 Nov 2026); [sentinelone.com](https://www.sentinelone.com/blog/prompt-security-for-agentic-ai/) (12 Nov 2026) | | Credential access | partial | ADR (early access, Oct 2026) says it blocks credential exfiltration before the action executes; endpoint file and credential access are not itemized. | [sentinelone.com](https://www.sentinelone.com/press/sentinelone-launches-singularity-agentic-detection-and-response-to-stop-rogue-and-malicious-ai-agents-at-machine-speed/) (12 Nov 2026) | | Data paste | documented | Policy rules redact, tokenize or block sensitive prompts. | [sentinelone.com](https://www.sentinelone.com/blog/a-new-chapter-for-ai-and-cybersecurity-sentinelone-acquires-prompt-security/) (12 Nov 2026); [sentinelone.com](https://www.sentinelone.com/platform/prompt-security-ai-usage-control/) (13 Nov 2026) | | Package install | none found | None found on 12 Nov 2026. | none found on 12 Nov 2026 | SentinelOne on the Action Control Matrix. SentinelOne: SentinelOne documents an MCP gateway with allow, block, filter or redact actions and runtime tool-call blocking, plus a lightweight agent and extension that discover shadow AI across browsers, IDEs and terminal assistants. Rows for every tool: [the Action Control Matrix](/action-control-matrix/). Method: [how a cell is read](/action-control-matrix/methodology/). All [six agent actions](/actions/), the [full index](/ranking/) and [every head-to-head](/compare/). ## Sources - www.sentinelone.com/blog/prompt-security-for-agentic-ai: https://www.sentinelone.com/blog/prompt-security-for-agentic-ai/ (read 12 Nov 2026) - www.sentinelone.com/press/sentinelone-expands-ai-security-coverage-to-claude-cowork: https://www.sentinelone.com/press/sentinelone-expands-ai-security-coverage-to-claude-cowork/ (read 12 Nov 2026) - www.sentinelone.com/press/sentinelone-launches-singularity-agentic-detection-and-response-to-stop-rogue-and-malicious-ai-agents-at-machine-speed: https://www.sentinelone.com/press/sentinelone-launches-singularity-agentic-detection-and-response-to-stop-rogue-and-malicious-ai-agents-at-machine-speed/ (read 12 Nov 2026) - www.sentinelone.com/blog/a-new-chapter-for-ai-and-cybersecurity-sentinelone-acquires-prompt-security: https://www.sentinelone.com/blog/a-new-chapter-for-ai-and-cybersecurity-sentinelone-acquires-prompt-security/ (read 12 Nov 2026) - www.sentinelone.com/platform/prompt-security-ai-usage-control: https://www.sentinelone.com/platform/prompt-security-ai-usage-control/ (read 13 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/action-control-matrix/sentinelone/ --- # WitnessAI can stop 2 of 6 agent actions, with 2 more in part DATASET ROW, read 13 Nov 2026 **The short answer.** WitnessAI ranks 11th of 16 on the index, and on the Action Control Matrix it has 2 documented cells, 2 partial and 2 none found on 13 Nov 2026. Each cell below carries a note, a source and a read date. Rows follow the site ranking. Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). ## WitnessAI, cell by cell File read none found; Shell command partial; Network call documented; Credential access partial; Data paste documented; Package install none found. | Action | Level | Note | Source and read date | | --- | --- | --- | --- | | File read | none found | No file-read blocking documented. | none found on 12 Nov 2026 | | Shell command | partial | Captures shell executions for supported agents; blocking of shell commands is not documented. | [witness.ai](https://witness.ai/for-agents/) (12 Nov 2026) | | Network call | documented | Blocks agent tool calls and MCP server connections inline in the traffic path. | [witness.ai](https://witness.ai/product/) (12 Nov 2026); [witness.ai](https://witness.ai/for-agents/) (12 Nov 2026) | | Credential access | partial | Tokenizes credentials and secrets before they reach a model or agent. | [witness.ai](https://witness.ai/protect/) (12 Nov 2026) | | Data paste | documented | Redacts sensitive data in prompts in real time. | [witness.ai](https://witness.ai/) (12 Nov 2026); [witness.ai](https://witness.ai/protect/) (12 Nov 2026) | | Package install | none found | No package-install blocking documented. | none found on 12 Nov 2026 | WitnessAI on the Action Control Matrix. WitnessAI: WitnessAI documents inline blocking of unapproved MCP servers and tool calls across IDEs, chat apps and custom agents, with audit records per blocked call and one policy for humans and agents. Rows for every tool: [the Action Control Matrix](/action-control-matrix/). Method: [how a cell is read](/action-control-matrix/methodology/). All [six agent actions](/actions/), the [full index](/ranking/) and [every head-to-head](/compare/). ## Sources - witness.ai/for-agents: https://witness.ai/for-agents/ (read 12 Nov 2026) - witness.ai/product: https://witness.ai/product/ (read 12 Nov 2026) - witness.ai/blog/witnessai-in-sacrs-arise-market-map-why-a-unified-ai-control-plane-matters: https://witness.ai/blog/witnessai-in-sacrs-arise-market-map-why-a-unified-ai-control-plane-matters/ (read 12 Nov 2026) - witness.ai/protect: https://witness.ai/protect/ (read 12 Nov 2026) - witness.ai: https://witness.ai/ (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/action-control-matrix/witnessai/ --- # Koi can stop 0 of 6 agent actions, with 4 more in part DATASET ROW, read 13 Nov 2026 **The short answer.** Koi ranks 12th of 16 on the index, and on the Action Control Matrix it has 0 documented cells, 4 partial and 2 none found on 13 Nov 2026. Each cell below carries a note, a source and a read date. Rows follow the site ranking. Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). ## Koi, cell by cell File read none found; Shell command partial; Network call partial; Credential access none found; Data paste partial; Package install partial. | Action | Level | Note | Source and read date | | --- | --- | --- | --- | | File read | none found | No file-read blocking documented. | [paloaltonetworks.com](https://www.paloaltonetworks.com/cortex/agentic-endpoint-security) (12 Nov 2026) | | Shell command | partial | Blocks rogue AI agent actions in general terms; shell commands not named. | [paloaltonetworks.com](https://www.paloaltonetworks.com/resources/datasheets/cortex-agentic-endpoint-security-at-a-glance) (12 Nov 2026); [paloaltonetworks.com](https://www.paloaltonetworks.com/cortex/agentic-endpoint-security) (12 Nov 2026) | | Network call | partial | Blocks malicious URLs via Prisma AIRS integration. | [paloaltonetworks.com](https://www.paloaltonetworks.com/cortex/agentic-endpoint-security.md) (12 Nov 2026) | | Credential access | none found | No credential-access blocking documented. | [paloaltonetworks.com](https://www.paloaltonetworks.com/cortex/agentic-endpoint-security) (12 Nov 2026) | | Data paste | partial | Blocks data exfiltration across AI interactions via Prisma AIRS. | [paloaltonetworks.com](https://www.paloaltonetworks.com/cortex/agentic-endpoint-security.md) (12 Nov 2026) | | Package install | partial | Stops risky downloads and offers one-click removal of insecure packages and extensions. | [paloaltonetworks.com](https://www.paloaltonetworks.com/resources/datasheets/cortex-agentic-endpoint-security-at-a-glance) (12 Nov 2026); [paloaltonetworks.com](https://www.paloaltonetworks.com/cortex/agentic-endpoint-security.md) (12 Nov 2026) | Koi on the Action Control Matrix. Koi: Palo Alto Networks documents inventory of IDE plugins, browser extensions and MCP servers on endpoints, plus real-time blocking of risky downloads and agent actions, with standalone or Cortex XDR deployment. Rows for every tool: [the Action Control Matrix](/action-control-matrix/). Method: [how a cell is read](/action-control-matrix/methodology/). All [six agent actions](/actions/), the [full index](/ranking/) and [every head-to-head](/compare/). ## Sources - www.paloaltonetworks.com/cortex/agentic-endpoint-security (checked, nothing found): https://www.paloaltonetworks.com/cortex/agentic-endpoint-security (read 12 Nov 2026) - www.paloaltonetworks.com/resources/datasheets/cortex-agentic-endpoint-security-at-a-glance: https://www.paloaltonetworks.com/resources/datasheets/cortex-agentic-endpoint-security-at-a-glance (read 12 Nov 2026) - www.paloaltonetworks.com/cortex/agentic-endpoint-security.md: https://www.paloaltonetworks.com/cortex/agentic-endpoint-security.md (read 12 Nov 2026) - www.calcalistech.com/ctechnews/article/xpz78znuc: https://www.calcalistech.com/ctechnews/article/xpz78znuc (read 12 Nov 2026) - www.securityweek.com/palo-alto-networks-to-acquire-koi-in-reported-400-million-transaction: https://www.securityweek.com/palo-alto-networks-to-acquire-koi-in-reported-400-million-transaction/ (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/action-control-matrix/koi/ --- # Lasso Security can stop 1 of 6 agent actions, with 3 more in part DATASET ROW, read 13 Nov 2026 **The short answer.** Lasso Security ranks 13th of 16 on the index, and on the Action Control Matrix it has 1 documented cells, 3 partial and 2 none found on 13 Nov 2026. Each cell below carries a note, a source and a read date. Rows follow the site ranking. Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). ## Lasso Security, cell by cell File read none found; Shell command partial; Network call partial; Credential access partial; Data paste documented; Package install none found. | Action | Level | Note | Source and read date | | --- | --- | --- | --- | | File read | none found | File reads not named as a blockable action. | [lasso.security](https://www.lasso.security/use-cases/ai-coding-assistants) (12 Nov 2026) | | Shell command | partial | Tool calls inspected before execution and flagged or blocked; shell not named as a distinct control. | [lasso.security](https://www.lasso.security/use-cases/ai-coding-assistants) (12 Nov 2026); [lasso.security](https://www.lasso.security/use-cases/ai-coding-assistants) (12 Nov 2026) | | Network call | partial | Blocks risky MCP servers and injection at the connection layer. | [lasso.security](https://www.lasso.security/use-cases/mcp-security) (12 Nov 2026); [github.com](https://github.com/lasso-security/kong-ai-lasso-guardrail) (12 Nov 2026) | | Credential access | partial | Credentials in MCP tool calls are detected and masked. | [lasso.security](https://www.lasso.security/use-cases/mcp-security) (12 Nov 2026) | | Data paste | documented | Prompts and uploads with PII or PCI are masked before reaching third-party models. | [lasso.security](https://www.lasso.security/platform/ai-usage-control) (12 Nov 2026) | | Package install | none found | No package install controls found. | [lasso.security](https://www.lasso.security/use-cases/ai-coding-assistants) (12 Nov 2026) | Lasso Security on the Action Control Matrix. Lasso Security: Lasso documents inspecting every coding-agent tool call before execution via Claude Code, Cursor and Codex hooks, with out-of-scope calls flagged or blocked, plus an open-source MCP gateway. Rows for every tool: [the Action Control Matrix](/action-control-matrix/). Method: [how a cell is read](/action-control-matrix/methodology/). All [six agent actions](/actions/), the [full index](/ranking/) and [every head-to-head](/compare/). ## Sources - www.lasso.security/use-cases/ai-coding-assistants (checked, nothing found): https://www.lasso.security/use-cases/ai-coding-assistants (read 12 Nov 2026) - www.lasso.security/use-cases/mcp-security: https://www.lasso.security/use-cases/mcp-security (read 12 Nov 2026) - github.com/lasso-security/kong-ai-lasso-guardrail: https://github.com/lasso-security/kong-ai-lasso-guardrail (read 12 Nov 2026) - www.lasso.security/platform/ai-usage-control: https://www.lasso.security/platform/ai-usage-control (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/action-control-matrix/lasso-security/ --- # Pillar Security can stop 4 of 6 agent actions, with 0 more in part DATASET ROW, read 13 Nov 2026 **The short answer.** Pillar Security ranks 14th of 16 on the index, and on the Action Control Matrix it has 4 documented cells, 0 partial and 2 none found on 13 Nov 2026. Each cell below carries a note, a source and a read date. Rows follow the site ranking. Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). ## Pillar Security, cell by cell File read documented; Shell command documented; Network call documented; Credential access documented; Data paste none found; Package install none found. | Action | Level | Note | Source and read date | | --- | --- | --- | --- | | File read | documented | Blocks reads of keys.env, SSH and cloud credentials from reaching external destinations. | [pillar.security](https://www.pillar.security/platform/agentic-endpoint) (12 Nov 2026) | | Shell command | documented | Pre-tool hook decisions before each action; risky commands blocked before execution per the page. | [pillar.security](https://www.pillar.security/platform/agentic-endpoint) (12 Nov 2026); [pillar.security](https://www.pillar.security/platform/agentic-endpoint) (12 Nov 2026) | | Network call | documented | Credential-read flows to external destinations blocked. | [pillar.security](https://www.pillar.security/platform/agentic-endpoint) (12 Nov 2026) | | Credential access | documented | Credential exfiltration blocking named explicitly. | [pillar.security](https://www.pillar.security/platform/agentic-endpoint) (12 Nov 2026); [pulse2.com](https://pulse2.com/vanta-to-acquire-pillar-security-to-expand-ai-agent-protection-as-customer-base-surpasses-20000/) (12 Nov 2026) | | Data paste | none found | Nothing found on 12 Nov 2026 on paste or upload controls for AI tools. | none found on 12 Nov 2026 | | Package install | none found | Nothing found on 12 Nov 2026 on package-install blocking. | none found on 12 Nov 2026 | Pillar Security on the Action Control Matrix. Pillar Security: Documents pre-action blocking in the agent's own pre-tool hook for Claude Code, Codex CLI and Cursor, with credential-exfiltration and persistence controls and MDM-installed hooks developers cannot disable. Rows for every tool: [the Action Control Matrix](/action-control-matrix/). Method: [how a cell is read](/action-control-matrix/methodology/). All [six agent actions](/actions/), the [full index](/ranking/) and [every head-to-head](/compare/). ## Sources - www.pillar.security/platform/agentic-endpoint: https://www.pillar.security/platform/agentic-endpoint (read 12 Nov 2026) - pulse2.com/vanta-to-acquire-pillar-security-to-expand-ai-agent-protection-as-customer-base-surpasses-20000: https://pulse2.com/vanta-to-acquire-pillar-security-to-expand-ai-agent-protection-as-customer-base-surpasses-20000/ (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/action-control-matrix/pillar-security/ --- # LayerX can stop 1 of 6 agent actions, with 1 more in part DATASET ROW, read 13 Nov 2026 **The short answer.** LayerX ranks 15th of 16 on the index, and on the Action Control Matrix it has 1 documented cells, 1 partial and 4 none found on 13 Nov 2026. Each cell below carries a note, a source and a read date. Rows follow the site ranking. Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). ## LayerX, cell by cell File read none found; Shell command none found; Network call none found; Credential access none found; Data paste documented; Package install partial. | Action | Level | Note | Source and read date | | --- | --- | --- | --- | | File read | none found | None found on 12 Nov 2026. | none found on 12 Nov 2026 | | Shell command | none found | None found on 12 Nov 2026. | none found on 12 Nov 2026 | | Network call | none found | None found on 12 Nov 2026: outbound call blocking outside the browser. | none found on 12 Nov 2026 | | Credential access | none found | None found on 12 Nov 2026. | none found on 12 Nov 2026 | | Data paste | documented | DLP over typed input, copy-paste and uploads with block or redact before data leaves. | [akamai.com](https://www.akamai.com/products/akamai-workforce-protector) (12 Nov 2026); [akamai.com](https://www.akamai.com/solutions/ai-usage-control) (12 Nov 2026) | | Package install | partial | Blocks malicious or overprivileged browser extensions; code package installs are not covered. | [akamai.com](https://www.akamai.com/products/akamai-workforce-protector) (12 Nov 2026) | LayerX on the Action Control Matrix. LayerX: It enforces warn, redact, block or guide policy on prompts, uploads and copy-paste at the moment of interaction with no proxy or browser replacement, and covers agentic browsers. Rows for every tool: [the Action Control Matrix](/action-control-matrix/). Method: [how a cell is read](/action-control-matrix/methodology/). All [six agent actions](/actions/), the [full index](/ranking/) and [every head-to-head](/compare/). ## Sources - www.akamai.com/products/akamai-workforce-protector: https://www.akamai.com/products/akamai-workforce-protector (read 12 Nov 2026) - www.akamai.com/solutions/ai-usage-control: https://www.akamai.com/solutions/ai-usage-control (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/action-control-matrix/layerx/ --- # Palo Alto Networks can stop 1 of 6 agent actions, with 2 more in part DATASET ROW, read 13 Nov 2026 **The short answer.** Palo Alto Networks ranks 16th of 16 on the index, and on the Action Control Matrix it has 1 documented cells, 2 partial and 3 none found on 13 Nov 2026. Each cell below carries a note, a source and a read date. Rows follow the site ranking. Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). ## Palo Alto Networks, cell by cell File read none found; Shell command partial; Network call partial; Credential access none found; Data paste documented; Package install none found. | Action | Level | Note | Source and read date | | --- | --- | --- | --- | | File read | none found | No file-read blocking documented. | [paloaltonetworks.com](https://www.paloaltonetworks.com/prisma/prisma-ai-runtime-security) (12 Nov 2026) | | Shell command | partial | Cortex XDR One guardrails are said to help stop destructive commands (Oct 2026); shell commands are not itemized and AES docs were not found. | [paloaltonetworks.com](https://www.paloaltonetworks.com/prisma/prisma-ai-runtime-security) (12 Nov 2026); [paloaltonetworks.com](https://www.paloaltonetworks.com/prisma/prisma-ai-runtime-security) (12 Nov 2026) | | Network call | partial | AIRS Network Intercept mode inspects AI traffic; blocking specifics not read. | [paloaltonetworks.com](https://www.paloaltonetworks.com/prisma/prisma-ai-runtime-security) (12 Nov 2026) | | Credential access | none found | No credential-access blocking documented. | [paloaltonetworks.com](https://www.paloaltonetworks.com/prisma/prisma-ai-runtime-security) (12 Nov 2026) | | Data paste | documented | Prisma Browser restricts copy-paste and uploads to unsanctioned AI apps and redacts prompts. | [paloaltonetworks.com](https://www.paloaltonetworks.com/sase/prisma-browser) (12 Nov 2026); [paloaltonetworks.com](https://www.paloaltonetworks.com/sase/prisma-browser) (12 Nov 2026) | | Package install | none found | None found on 12 Nov 2026 for Prisma AIRS or Prisma Browser; Cortex Agentic Endpoint Security (Koi) coverage is scored on the Koi profile. | none found on 12 Nov 2026 | Palo Alto Networks on the Action Control Matrix. Palo Alto Networks: Palo Alto Networks documents both a browser-level control point (extension inventory, copy-paste and upload limits for unsanctioned AI apps) and endpoint coding-agent policy enforcement under Prisma AIRS, backed by public docs and analyst recognition. Rows for every tool: [the Action Control Matrix](/action-control-matrix/). Method: [how a cell is read](/action-control-matrix/methodology/). All [six agent actions](/actions/), the [full index](/ranking/) and [every head-to-head](/compare/). ## Sources - www.paloaltonetworks.com/prisma/prisma-ai-runtime-security (checked, nothing found): https://www.paloaltonetworks.com/prisma/prisma-ai-runtime-security (read 12 Nov 2026) - www.paloaltonetworks.com/blog/2026/10/reinventing-endpoint-security-with-cortex-xdr-one: https://www.paloaltonetworks.com/blog/2026/10/reinventing-endpoint-security-with-cortex-xdr-one/ (read 12 Nov 2026) - www.paloaltonetworks.com/sase/prisma-browser: https://www.paloaltonetworks.com/sase/prisma-browser (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/action-control-matrix/palo-alto-networks/ --- # 9 of 16 tools document coverage of desktop apps AGENT SURFACE, read 13 Nov 2026 **The short answer.** 9 of 16 AI endpoint security tools document full coverage of desktop apps, 6 document it in part, and nothing was found for 1 on 13 Nov 2026. Notes and sources for all 16 tools sit in the table below. Rows are ordered by documentation level, then by the site ranking. ## Which tools see desktop AI apps? 9 of 16 tools document full coverage of desktop apps: Microsoft, Harmonic Security, Aim Security, CrowdStrike, Island, Knostic, SentinelOne, WitnessAI and LayerX. 6 more do in part (Bay, Noma Security, Zenity, Koi, Lasso Security and Palo Alto Networks). ## What this surface is Chat and agent applications installed on a laptop, such as a desktop assistant that can read local folders or run tools on the user's behalf. ## Tool by tool | Tool | Level | Note | Source | | --- | --- | --- | --- | | [Microsoft](/platforms/microsoft/) | documented | Discovers Claude Desktop, ChatGPT Desktop, Codex Desktop and others; network inspection covers ChatGPT Desktop. | [learn.microsoft.com](https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-support-matrix), 12 Nov 2026 | | [Harmonic Security](/platforms/harmonic-security/) | documented | Discovery names Claude Desktop and ChatGPT Desktop; inline prompt controls are early access and Windows only. | [harmonic.security](https://www.harmonic.security/solutions/endpoint-ai-security), 12 Nov 2026 | | [Aim Security](/platforms/aim-security/) | documented | Desktop AI apps are covered by the endpoint client and prompts are checked at send. | [catonetworks.com](https://www.catonetworks.com/platform/cato-ai-endpoint/), 12 Nov 2026 | | [CrowdStrike](/platforms/crowdstrike/) | documented | A third party reports AIDR for Endpoint covers desktop AI apps; Guardian discovers agents on endpoints and blocks sensitive data in prompts. | [msspalert.com](https://www.msspalert.com/news/crowdstrike-brings-ai-security-to-the-endpoint-and-the-timing-makes-sense), 12 Nov 2026 | | [Island](/platforms/island/) | documented | Island Desktop applies identity-aware policy and DLP to native apps and files on Windows and macOS; desktop agents are listed as governed. | [island.io](https://www.island.io/enterprise-desktop), 12 Nov 2026 | | [Knostic](/platforms/knostic/) | documented | Claude Desktop and Cowork supported via endpoint client and plugin. | [knostic.ai](https://www.knostic.ai/faq), 12 Nov 2026 | | [SentinelOne](/platforms/sentinelone/) | documented | Agent discovers sanctioned and shadow AI on desktop apps; Claude Cowork VM agent announced. | [sentinelone.com](https://www.sentinelone.com/blog/a-new-chapter-for-ai-and-cybersecurity-sentinelone-acquires-prompt-security/), 12 Nov 2026 | | [WitnessAI](/platforms/witnessai/) | documented | Says it covers desktop clients and native apps such as Windows 11 Copilot, via network traffic. | [witness.ai](https://witness.ai/for-agents/), 12 Nov 2026 | | [LayerX](/platforms/layerx/) | documented | LayerX says a native desktop agent enforces policy on ChatGPT desktop, Claude desktop and Microsoft 365 Copilot desktop (vendor release, May 2026). | [akamai.com](https://www.akamai.com/solutions/ai-usage-control), 12 Nov 2026 | | [Bay](/platforms/bay/) | partial | Bay says it finds agents and models on employee devices and names Claude Code and Cursor; desktop chat apps are not named on its pages. | [bay.io](https://www.bay.io/platform/ai-endpoint-security?utm_source=aiendpointsecurityranked.com&utm_medium=referral&utm_campaign=surface), 12 Nov 2026 | | [Noma Security](/platforms/noma-security/) | partial | Claude Cowork named as an endpoint agent; desktop app inventory method not itemized. | [noma.security](https://noma.security/platform/endpoint-agents), 12 Nov 2026 | | [Zenity](/platforms/zenity/) | partial | Cowork and Claude Enterprise desktop named on the coding and personal agents page; no inventory method detailed. | [zenity.io](https://zenity.io/use-cases/agent-type/coding-personal-agents), 12 Nov 2026 | | [Koi](/platforms/koi/) | partial | Lists user-deployed AI applications as governed endpoint items, without naming ChatGPT or Claude desktop. | [paloaltonetworks.com](https://www.paloaltonetworks.com/resources/techbriefs/koi-agentic-endpoint-security), 12 Nov 2026 | | [Lasso Security](/platforms/lasso-security/) | partial | Says it discovers local desktop agents; no named desktop apps or enforcement detail. | [lasso.security](https://www.lasso.security/platform/ai-usage-control), 12 Nov 2026 | | [Palo Alto Networks](/platforms/palo-alto-networks/) | partial | Prisma AIRS says it discovers AI apps and agents; desktop apps are not named. | [paloaltonetworks.com](https://www.paloaltonetworks.com/prisma/prisma-ai-runtime-security), 12 Nov 2026 | | [Pillar Security](/platforms/pillar-security/) | none found | Nothing found on 12 Nov 2026 for desktop AI apps such as ChatGPT or Claude desktop. | none found on 12 Nov 2026 | Ordered by documentation level, then by the site ranking. Read 13 Nov 2026. Related: [Agent Surface Map](/agent-surface-map/), [how a cell is read](/agent-surface-map/methodology/). **About Bay.** Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). ## The other agent surfaces - [IDE assistants](/surfaces/ide-assistants/) - [Command-line agents](/surfaces/cli-agents/) - [MCP servers](/surfaces/mcp-servers/) - [Browser extensions](/surfaces/browser-extensions/) - [SaaS copilots](/surfaces/saas-copilots/) ## Sources - learn.microsoft.com/en-us/defender-endpoint/ai-agent-support-matrix: https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-support-matrix (read 12 Nov 2026) - learn.microsoft.com/en-us/microsoft-365/admin/manage/agent-shadow-ai: https://learn.microsoft.com/en-us/microsoft-365/admin/manage/agent-shadow-ai (read 12 Nov 2026) - www.harmonic.security/solutions/endpoint-ai-security: https://www.harmonic.security/solutions/endpoint-ai-security (read 12 Nov 2026) - docs.harmonicsecurity.app/portal-guides/detection-and-response/sensitive-data-controls-for-coding-agents-and-desktop-applications: https://docs.harmonicsecurity.app/portal-guides/detection-and-response/sensitive-data-controls-for-coding-agents-and-desktop-applications (read 12 Nov 2026) - www.catonetworks.com/platform/cato-ai-endpoint: https://www.catonetworks.com/platform/cato-ai-endpoint/ (read 12 Nov 2026) - www.msspalert.com/news/crowdstrike-brings-ai-security-to-the-endpoint-and-the-timing-makes-sense: https://www.msspalert.com/news/crowdstrike-brings-ai-security-to-the-endpoint-and-the-timing-makes-sense (read 12 Nov 2026) - www.crowdstrike.com/en-us/press-releases/crowdstrike-announces-general-availability-of-falcon-ai-detection-and-response: https://www.crowdstrike.com/en-us/press-releases/crowdstrike-announces-general-availability-of-falcon-ai-detection-and-response/ (read 12 Nov 2026) - www.island.io/enterprise-desktop: https://www.island.io/enterprise-desktop (read 12 Nov 2026) - www.knostic.ai/faq: https://www.knostic.ai/faq (read 12 Nov 2026) - www.sentinelone.com/blog/a-new-chapter-for-ai-and-cybersecurity-sentinelone-acquires-prompt-security: https://www.sentinelone.com/blog/a-new-chapter-for-ai-and-cybersecurity-sentinelone-acquires-prompt-security/ (read 12 Nov 2026) - www.sentinelone.com/press/sentinelone-expands-ai-security-coverage-to-claude-cowork: https://www.sentinelone.com/press/sentinelone-expands-ai-security-coverage-to-claude-cowork/ (read 12 Nov 2026) - witness.ai/for-agents: https://witness.ai/for-agents/ (read 12 Nov 2026) - witness.ai/product: https://witness.ai/product/ (read 12 Nov 2026) - www.akamai.com/solutions/ai-usage-control: https://www.akamai.com/solutions/ai-usage-control (read 12 Nov 2026) - www.manilatimes.net/2026/05/21/tmt-newswire/globenewswire/layerx-security-integrates-with-new-extended-plan-for-aws-security-hub/2348650: https://www.manilatimes.net/2026/05/21/tmt-newswire/globenewswire/layerx-security-integrates-with-new-extended-plan-for-aws-security-hub/2348650 (read 13 Nov 2026) - www.bay.io/platform/ai-endpoint-security: https://www.bay.io/platform/ai-endpoint-security (read 12 Nov 2026) - noma.security/platform/endpoint-agents: https://noma.security/platform/endpoint-agents (read 12 Nov 2026) - noma.security/blog/noma-brings-agent-security-to-every-employee-endpoint: https://noma.security/blog/noma-brings-agent-security-to-every-employee-endpoint (read 12 Nov 2026) - zenity.io/use-cases/agent-type/coding-personal-agents: https://zenity.io/use-cases/agent-type/coding-personal-agents (read 12 Nov 2026) - www.paloaltonetworks.com/resources/techbriefs/koi-agentic-endpoint-security: https://www.paloaltonetworks.com/resources/techbriefs/koi-agentic-endpoint-security (read 12 Nov 2026) - www.lasso.security/platform/ai-usage-control: https://www.lasso.security/platform/ai-usage-control (read 12 Nov 2026) - www.lasso.security/blog/securing-ai-on-endpoints-how-to-govern-ai-tools-coding-agents-and-mcp-connections: https://www.lasso.security/blog/securing-ai-on-endpoints-how-to-govern-ai-tools-coding-agents-and-mcp-connections (read 12 Nov 2026) - www.paloaltonetworks.com/prisma/prisma-ai-runtime-security: https://www.paloaltonetworks.com/prisma/prisma-ai-runtime-security (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/surfaces/desktop-apps/ --- # 14 of 16 tools document coverage of IDE assistants AGENT SURFACE, read 13 Nov 2026 **The short answer.** 14 of 16 AI endpoint security tools document full coverage of IDE assistants, 1 document it in part, and nothing was found for 1 on 13 Nov 2026. Notes and sources for all 16 tools sit in the table below. Rows are ordered by documentation level, then by the site ranking. ## Which tools see IDE assistants? 14 of 16 tools document full coverage of IDE assistants: Microsoft, Bay, Harmonic Security, CrowdStrike, Noma Security, Island, Zenity, Knostic, SentinelOne, WitnessAI, Koi, Lasso Security, Pillar Security and Palo Alto Networks. 1 more does in part (Aim Security). ## What this surface is Assistants inside code editors and IDEs, which read the open repository and can propose or apply edits and commands. ## Tool by tool | Tool | Level | Note | Source | | --- | --- | --- | --- | | [Microsoft](/platforms/microsoft/) | documented | Discovers Cursor, Windsurf and VS Code assistant extensions; hooks cover GitHub Copilot extension. | [learn.microsoft.com](https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-support-matrix), 12 Nov 2026 | | [Bay](/platforms/bay/) | documented | Bay names Cursor, and Software Analyst notes IDE extensions in the entity graph. | [bay.io](https://www.bay.io/platform/ai-endpoint-security?utm_source=aiendpointsecurityranked.com&utm_medium=referral&utm_campaign=surface), 12 Nov 2026 | | [Harmonic Security](/platforms/harmonic-security/) | documented | Cursor, VS Code and Copilot covered by discovery and hook-based prompt and tool-call controls. | [harmonic.security](https://www.harmonic.security/solutions/endpoint-ai-security), 12 Nov 2026 | | [CrowdStrike](/platforms/crowdstrike/) | documented | Hooks cover Cursor and GitHub Copilot; discovery lists IDEs. | [crowdstrike.com](https://www.crowdstrike.com/en-us/blog/crowdstrike-advances-ai-security-and-safety-falcon-guardian/), 12 Nov 2026 | | [Noma Security](/platforms/noma-security/) | documented | IDEs and Cursor, Windsurf, Kiro named; Cursor hooks integration documented. | [noma.security](https://noma.security/blog/noma-brings-agent-security-to-every-employee-endpoint), 12 Nov 2026 | | [Island](/platforms/island/) | documented | Names Cursor, Windsurf and Cline and inventories IDE extensions and flags over-permissioned ones. | [island.io](https://www.island.io/product/ai), 12 Nov 2026 | | [Zenity](/platforms/zenity/) | documented | Cursor and GitHub Copilot named; configuration and hook telemetry documented. | [zenity.io](https://zenity.io/use-cases/agent-type/coding-personal-agents), 12 Nov 2026 | | [Knostic](/platforms/knostic/) | documented | Cursor, Windsurf, VS Code Copilot via IDE extension. | [knostic.ai](https://www.knostic.ai/faq), 12 Nov 2026 | | [SentinelOne](/platforms/sentinelone/) | documented | Desktop IDEs and embedded agents in developer tools are named in discovery. | [sentinelone.com](https://www.sentinelone.com/blog/a-new-chapter-for-ai-and-cybersecurity-sentinelone-acquires-prompt-security/), 12 Nov 2026 | | [WitnessAI](/platforms/witnessai/) | documented | Captures prompts and tool calls across supported Cursor, Claude Code and Codex workflows and IDEs. | [witness.ai](https://witness.ai/for-agents/), 12 Nov 2026 | | [Koi](/platforms/koi/) | documented | Discovers IDE plugins and extensions on endpoints. | [paloaltonetworks.com](https://www.paloaltonetworks.com/resources/datasheets/cortex-agentic-endpoint-security-at-a-glance), 12 Nov 2026 | | [Lasso Security](/platforms/lasso-security/) | documented | Cursor and other coding assistants covered through hooks. | [lasso.security](https://www.lasso.security/use-cases/ai-coding-assistants), 12 Nov 2026 | | [Pillar Security](/platforms/pillar-security/) | documented | Cursor deep scan and pre-action hook documented; Antigravity and others by fingerprint. | [pillar.security](https://www.pillar.security/platform/agentic-endpoint), 12 Nov 2026 | | [Palo Alto Networks](/platforms/palo-alto-networks/) | documented | AES discovers IDE plugins; AIRS release covers AI coding agents on endpoints. | [paloaltonetworks.com](https://www.paloaltonetworks.com/prisma/prisma-ai-runtime-security), 12 Nov 2026 | | [Aim Security](/platforms/aim-security/) | partial | Cursor is named as a covered coding agent; no IDE-extension inventory documented. | [catonetworks.com](https://www.catonetworks.com/platform/cato-ai-endpoint/), 12 Nov 2026 | | [LayerX](/platforms/layerx/) | none found | None found on 12 Nov 2026: product coverage of IDE assistants; only research on Cursor extensions exists. | none found on 12 Nov 2026 | Ordered by documentation level, then by the site ranking. Read 13 Nov 2026. Related: [Agent Surface Map](/agent-surface-map/), [how a cell is read](/agent-surface-map/methodology/). **About Bay.** Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). ## The other agent surfaces - [Desktop apps](/surfaces/desktop-apps/) - [Command-line agents](/surfaces/cli-agents/) - [MCP servers](/surfaces/mcp-servers/) - [Browser extensions](/surfaces/browser-extensions/) - [SaaS copilots](/surfaces/saas-copilots/) ## Sources - learn.microsoft.com/en-us/defender-endpoint/ai-agent-support-matrix: https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-support-matrix (read 12 Nov 2026) - www.bay.io/platform/ai-endpoint-security: https://www.bay.io/platform/ai-endpoint-security (read 12 Nov 2026) - softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control: https://softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control (read 12 Nov 2026) - www.harmonic.security/solutions/endpoint-ai-security: https://www.harmonic.security/solutions/endpoint-ai-security (read 12 Nov 2026) - docs.harmonicsecurity.app/portal-guides/detection-and-response/sensitive-data-controls-for-coding-agents-and-desktop-applications: https://docs.harmonicsecurity.app/portal-guides/detection-and-response/sensitive-data-controls-for-coding-agents-and-desktop-applications (read 12 Nov 2026) - www.crowdstrike.com/en-us/blog/crowdstrike-advances-ai-security-and-safety-falcon-guardian: https://www.crowdstrike.com/en-us/blog/crowdstrike-advances-ai-security-and-safety-falcon-guardian/ (read 12 Nov 2026) - www.crowdstrike.com/en-us/blog/crowdstrike-shadow-AI-visibility-service: https://www.crowdstrike.com/en-us/blog/crowdstrike-shadow-AI-visibility-service/ (read 12 Nov 2026) - noma.security/blog/noma-brings-agent-security-to-every-employee-endpoint: https://noma.security/blog/noma-brings-agent-security-to-every-employee-endpoint (read 12 Nov 2026) - noma.security/blog/securing-the-agentic-frontier-noma-unveils-the-first-real-time-agent-runtime-security-for-cursor: https://noma.security/blog/securing-the-agentic-frontier-noma-unveils-the-first-real-time-agent-runtime-security-for-cursor (read 12 Nov 2026) - github.com/Noma-Security/noma-marketplace: https://github.com/Noma-Security/noma-marketplace (read 12 Nov 2026) - www.island.io/product/ai: https://www.island.io/product/ai (read 12 Nov 2026) - www.island.io/blog/expanding-island-to-govern-your-ai-workforce: https://www.island.io/blog/expanding-island-to-govern-your-ai-workforce (read 12 Nov 2026) - zenity.io/use-cases/agent-type/coding-personal-agents: https://zenity.io/use-cases/agent-type/coding-personal-agents (read 12 Nov 2026) - zenity.io/blog/coding-agent-attack-surface: https://zenity.io/blog/coding-agent-attack-surface (read 12 Nov 2026) - www.knostic.ai/faq: https://www.knostic.ai/faq (read 12 Nov 2026) - www.knostic.ai/kirin: https://www.knostic.ai/kirin (read 12 Nov 2026) - www.sentinelone.com/blog/a-new-chapter-for-ai-and-cybersecurity-sentinelone-acquires-prompt-security: https://www.sentinelone.com/blog/a-new-chapter-for-ai-and-cybersecurity-sentinelone-acquires-prompt-security/ (read 12 Nov 2026) - www.sentinelone.com/blog/prompt-security-for-agentic-ai: https://www.sentinelone.com/blog/prompt-security-for-agentic-ai/ (read 12 Nov 2026) - www.sentinelone.com/platform/prompt-security-ai-usage-control: https://www.sentinelone.com/platform/prompt-security-ai-usage-control/ (read 13 Nov 2026) - witness.ai/for-agents: https://witness.ai/for-agents/ (read 12 Nov 2026) - www.paloaltonetworks.com/resources/datasheets/cortex-agentic-endpoint-security-at-a-glance: https://www.paloaltonetworks.com/resources/datasheets/cortex-agentic-endpoint-security-at-a-glance (read 12 Nov 2026) - www.paloaltonetworks.com/resources/techbriefs/koi-agentic-endpoint-security: https://www.paloaltonetworks.com/resources/techbriefs/koi-agentic-endpoint-security (read 12 Nov 2026) - www.lasso.security/use-cases/ai-coding-assistants: https://www.lasso.security/use-cases/ai-coding-assistants (read 12 Nov 2026) - www.pillar.security/platform/agentic-endpoint: https://www.pillar.security/platform/agentic-endpoint (read 12 Nov 2026) - www.paloaltonetworks.com/prisma/prisma-ai-runtime-security: https://www.paloaltonetworks.com/prisma/prisma-ai-runtime-security (read 12 Nov 2026) - www.catonetworks.com/platform/cato-ai-endpoint: https://www.catonetworks.com/platform/cato-ai-endpoint/ (read 12 Nov 2026) - www.catonetworks.com/news/cato-acquires-aim-security-to-extend-sase-leadership-and-secure-enterprise-ai-transformation: https://www.catonetworks.com/news/cato-acquires-aim-security-to-extend-sase-leadership-and-secure-enterprise-ai-transformation/ (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/surfaces/ide-assistants/ --- # 12 of 16 tools document coverage of command-line agents AGENT SURFACE, read 13 Nov 2026 **The short answer.** 12 of 16 AI endpoint security tools document full coverage of command-line agents, 3 document it in part, and nothing was found for 1 on 13 Nov 2026. Notes and sources for all 16 tools sit in the table below. Rows are ordered by documentation level, then by the site ranking. ## Which tools see command-line agents? 12 of 16 tools document full coverage of command-line agents: Microsoft, Bay, Harmonic Security, Aim Security, CrowdStrike, Noma Security, Zenity, Knostic, SentinelOne, Koi, Lasso Security and Pillar Security. 3 more do in part (Island, WitnessAI and Palo Alto Networks). ## What this surface is Coding agents that run in a terminal with the user's shell, files and credentials, usually started by a developer. ## Tool by tool | Tool | Level | Note | Source | | --- | --- | --- | --- | | [Microsoft](/platforms/microsoft/) | documented | Discovers Claude Code, Codex CLI, Gemini CLI; runtime protection covers Claude Code and Copilot CLI. | [learn.microsoft.com](https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-support-matrix), 12 Nov 2026 | | [Bay](/platforms/bay/) | documented | Bay names Claude Code, Codex, Gemini CLI and GitHub Copilot, and lists CLI tools in its inventory. | [bay.io](https://www.bay.io/platform/ai-endpoint-security?utm_source=aiendpointsecurityranked.com&utm_medium=referral&utm_campaign=surface), 12 Nov 2026 | | [Harmonic Security](/platforms/harmonic-security/) | documented | Claude Code, Codex CLI, Copilot CLI, Kiro CLI hooks documented; Grok Build blocks tool calls only. | [harmonic.security](https://www.harmonic.security/solutions/endpoint-ai-security), 12 Nov 2026 | | [Aim Security](/platforms/aim-security/) | documented | Local agents such as Claude Code and Codex are shown in the endpoint dashboard with MCP servers and skills. | [catonetworks.com](https://www.catonetworks.com/platform/cato-ai-endpoint/), 12 Nov 2026 | | [CrowdStrike](/platforms/crowdstrike/) | documented | Hooks cover Claude Code, OpenAI Codex and OpenCode; Linux discovery for Claude Code and Codex is pre-beta. | [crowdstrike.com](https://www.crowdstrike.com/en-us/blog/crowdstrike-advances-ai-security-and-safety-falcon-guardian/), 12 Nov 2026 | | [Noma Security](/platforms/noma-security/) | documented | Claude Code and Codex named on the endpoint page. | [noma.security](https://noma.security/platform/endpoint-agents), 12 Nov 2026 | | [Zenity](/platforms/zenity/) | documented | Claude Code, Codex named with permission-mode and hook behavior. | [zenity.io](https://zenity.io/use-cases/agent-type/coding-personal-agents), 12 Nov 2026 | | [Knostic](/platforms/knostic/) | documented | Claude Code CLI supported; Codex and Gemini CLI listed on Kirin page, roadmap in FAQ. | [knostic.ai](https://www.knostic.ai/faq), 12 Nov 2026 | | [SentinelOne](/platforms/sentinelone/) | documented | Terminal-based assistants are named in discovery, and the AI Usage Control page names Claude Code as tracked. | [sentinelone.com](https://www.sentinelone.com/blog/a-new-chapter-for-ai-and-cybersecurity-sentinelone-acquires-prompt-security/), 12 Nov 2026 | | [Koi](/platforms/koi/) | documented | Names Claude Code and OpenClaw as the agent tools it addresses; no CLI agent inventory detail found. | [paloaltonetworks.com](https://www.paloaltonetworks.com/company/press/2026/palo-alto-networks-completes-acquisition-of-koi-to-secure-the-agentic-endpoint), 12 Nov 2026 | | [Lasso Security](/platforms/lasso-security/) | documented | Claude Code, Codex and OpenCode covered through lifecycle hooks. | [lasso.security](https://www.lasso.security/use-cases/ai-coding-assistants), 12 Nov 2026 | | [Pillar Security](/platforms/pillar-security/) | documented | Claude Code, Codex CLI full runtime control; Gemini CLI, OpenCode and Pi scanned. | [pillar.security](https://www.pillar.security/platform/agentic-endpoint), 12 Nov 2026 | | [Island](/platforms/island/) | partial | Claude Code sessions are read via the Compliance API and Bash commands are inspected; other CLI agents are not named. | [island.io](https://www.island.io/blog/claude-code-cowork-and-claude-for-microsoft-365-now-report-into-island), 12 Nov 2026 | | [WitnessAI](/platforms/witnessai/) | partial | Claude Code and Codex workflows are named as supported; coverage depends on traffic passing through the platform. | [witness.ai](https://witness.ai/for-agents/), 12 Nov 2026 | | [Palo Alto Networks](/platforms/palo-alto-networks/) | partial | Coding-agent policy enforcement on endpoints is announced; CLI agents are not named on the page. | [paloaltonetworks.com](https://www.paloaltonetworks.com/prisma/prisma-ai-runtime-security), 12 Nov 2026 | | [LayerX](/platforms/layerx/) | none found | None found on 12 Nov 2026: coverage of Claude Code, Codex CLI or Gemini CLI as a product feature. | none found on 12 Nov 2026 | Ordered by documentation level, then by the site ranking. Read 13 Nov 2026. Related: [Agent Surface Map](/agent-surface-map/), [how a cell is read](/agent-surface-map/methodology/). **About Bay.** Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). ## The other agent surfaces - [Desktop apps](/surfaces/desktop-apps/) - [IDE assistants](/surfaces/ide-assistants/) - [MCP servers](/surfaces/mcp-servers/) - [Browser extensions](/surfaces/browser-extensions/) - [SaaS copilots](/surfaces/saas-copilots/) ## Sources - learn.microsoft.com/en-us/defender-endpoint/ai-agent-support-matrix: https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-support-matrix (read 12 Nov 2026) - learn.microsoft.com/en-us/defender-endpoint/ai-agent-runtime-protection-overview: https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-runtime-protection-overview (read 12 Nov 2026) - www.bay.io/platform/ai-endpoint-security: https://www.bay.io/platform/ai-endpoint-security (read 12 Nov 2026) - www.bay.io/platform/cloud-workspaces: https://www.bay.io/platform/cloud-workspaces (read 12 Nov 2026) - www.harmonic.security/solutions/endpoint-ai-security: https://www.harmonic.security/solutions/endpoint-ai-security (read 12 Nov 2026) - docs.harmonicsecurity.app/portal-guides/detection-and-response/sensitive-data-controls-for-coding-agents-and-desktop-applications: https://docs.harmonicsecurity.app/portal-guides/detection-and-response/sensitive-data-controls-for-coding-agents-and-desktop-applications (read 12 Nov 2026) - www.catonetworks.com/platform/cato-ai-endpoint: https://www.catonetworks.com/platform/cato-ai-endpoint/ (read 12 Nov 2026) - www.crowdstrike.com/en-us/blog/crowdstrike-advances-ai-security-and-safety-falcon-guardian: https://www.crowdstrike.com/en-us/blog/crowdstrike-advances-ai-security-and-safety-falcon-guardian/ (read 12 Nov 2026) - www.crowdstrike.com/en-us/blog/falcon-guardian-defines-next-generation-of-ai-security: https://www.crowdstrike.com/en-us/blog/falcon-guardian-defines-next-generation-of-ai-security/ (read 12 Nov 2026) - noma.security/platform/endpoint-agents: https://noma.security/platform/endpoint-agents (read 12 Nov 2026) - noma.security/blog/noma-brings-agent-security-to-every-employee-endpoint: https://noma.security/blog/noma-brings-agent-security-to-every-employee-endpoint (read 12 Nov 2026) - github.com/Noma-Security/noma-marketplace: https://github.com/Noma-Security/noma-marketplace (read 12 Nov 2026) - zenity.io/use-cases/agent-type/coding-personal-agents: https://zenity.io/use-cases/agent-type/coding-personal-agents (read 12 Nov 2026) - zenity.io/blog/coding-agent-attack-surface: https://zenity.io/blog/coding-agent-attack-surface (read 12 Nov 2026) - www.knostic.ai/faq: https://www.knostic.ai/faq (read 12 Nov 2026) - www.knostic.ai/kirin: https://www.knostic.ai/kirin (read 12 Nov 2026) - www.sentinelone.com/blog/a-new-chapter-for-ai-and-cybersecurity-sentinelone-acquires-prompt-security: https://www.sentinelone.com/blog/a-new-chapter-for-ai-and-cybersecurity-sentinelone-acquires-prompt-security/ (read 12 Nov 2026) - www.sentinelone.com/blog/prompt-security-for-agentic-ai: https://www.sentinelone.com/blog/prompt-security-for-agentic-ai/ (read 12 Nov 2026) - www.sentinelone.com/platform/prompt-security-ai-usage-control: https://www.sentinelone.com/platform/prompt-security-ai-usage-control/ (read 13 Nov 2026) - www.paloaltonetworks.com/company/press/2026/palo-alto-networks-completes-acquisition-of-koi-to-secure-the-agentic-endpoint: https://www.paloaltonetworks.com/company/press/2026/palo-alto-networks-completes-acquisition-of-koi-to-secure-the-agentic-endpoint (read 12 Nov 2026) - www.paloaltonetworks.com/blog/2026/02/securing-the-agentic-endpoint: https://www.paloaltonetworks.com/blog/2026/02/securing-the-agentic-endpoint/ (read 12 Nov 2026) - www.sdxcentral.com/news/palo-alto-networks-bolsters-ai-security-for-the-endpoint: https://www.sdxcentral.com/news/palo-alto-networks-bolsters-ai-security-for-the-endpoint/ (read 13 Nov 2026) - cortex-docs.paloaltonetworks.com/cortex-xdr-6.x-rn/readme/features-introduced-in-2026/november-2026/release-highlights.md: https://cortex-docs.paloaltonetworks.com/cortex-xdr-6.x-rn/readme/features-introduced-in-2026/november-2026/release-highlights.md (read 13 Nov 2026) - www.lasso.security/use-cases/ai-coding-assistants: https://www.lasso.security/use-cases/ai-coding-assistants (read 12 Nov 2026) - www.pillar.security/platform/agentic-endpoint: https://www.pillar.security/platform/agentic-endpoint (read 12 Nov 2026) - www.island.io/blog/claude-code-cowork-and-claude-for-microsoft-365-now-report-into-island: https://www.island.io/blog/claude-code-cowork-and-claude-for-microsoft-365-now-report-into-island (read 12 Nov 2026) - www.island.io/product/ai: https://www.island.io/product/ai (read 12 Nov 2026) - witness.ai/for-agents: https://witness.ai/for-agents/ (read 12 Nov 2026) - www.paloaltonetworks.com/prisma/prisma-ai-runtime-security: https://www.paloaltonetworks.com/prisma/prisma-ai-runtime-security (read 12 Nov 2026) - www.paloaltonetworks.com/blog/2026/10/reinventing-endpoint-security-with-cortex-xdr-one: https://www.paloaltonetworks.com/blog/2026/10/reinventing-endpoint-security-with-cortex-xdr-one/ (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/surfaces/cli-agents/ --- # 14 of 16 tools document coverage of MCP servers AGENT SURFACE, read 13 Nov 2026 **The short answer.** 14 of 16 AI endpoint security tools document full coverage of MCP servers, 0 document it in part, and nothing was found for 2 on 13 Nov 2026. Notes and sources for all 16 tools sit in the table below. Rows are ordered by documentation level, then by the site ranking. ## Which tools see MCP servers on a laptop? 14 of 16 tools document full coverage of MCP servers: Microsoft, Bay, Harmonic Security, Aim Security, CrowdStrike, Noma Security, Island, Zenity, Knostic, SentinelOne, WitnessAI, Koi, Lasso Security and Pillar Security. None covers it in part. ## What this surface is Model Context Protocol servers running on or reached from the endpoint, which hand agents tools, files and data sources. ## Tool by tool | Tool | Level | Note | Source | | --- | --- | --- | --- | | [Microsoft](/platforms/microsoft/) | documented | Discovers configured local and remote MCP servers per agent. | [learn.microsoft.com](https://learn.microsoft.com/en-us/defender-endpoint/local-agent-discovery-overview), 12 Nov 2026 | | [Bay](/platforms/bay/) | documented | Bay inventories MCP servers and can scope which servers may read or write. | [bay.io](https://www.bay.io/platform/ai-endpoint-security?utm_source=aiendpointsecurityranked.com&utm_medium=referral&utm_campaign=surface), 12 Nov 2026 | | [Harmonic Security](/platforms/harmonic-security/) | documented | Discovery plus local MCP gateway with block on unpermitted servers and tool controls. | [harmonic.security](https://www.harmonic.security/pricing), 12 Nov 2026 | | [Aim Security](/platforms/aim-security/) | documented | Finds MCP servers and skills per agent and profiles MCP servers and tool calls. | [catonetworks.com](https://www.catonetworks.com/platform/cato-ai-endpoint/), 12 Nov 2026 | | [CrowdStrike](/platforms/crowdstrike/) | documented | Discovers MCP servers and malicious MCP servers and skills. | [crowdstrike.com](https://www.crowdstrike.com/en-us/platform/falcon-guardian-aidr/), 12 Nov 2026 | | [Noma Security](/platforms/noma-security/) | documented | Approved and shadow MCP server discovery with tool exposure. | [noma.security](https://noma.security/solutions/mcp-security), 12 Nov 2026 | | [Island](/platforms/island/) | documented | Inventories MCP servers with risk scores and gateways MCP calls with policy and audit. | [island.io](https://www.island.io/product/ai), 12 Nov 2026 | | [Zenity](/platforms/zenity/) | documented | MCP inventory and tool-call governance, delivered via gateway and hooks. | [zenity.io](https://zenity.io/blog/coding-agent-attack-surface), 12 Nov 2026 | | [Knostic](/platforms/knostic/) | documented | MCP scanning, cross-org visibility and quarantine of risky servers. | [knostic.ai](https://www.knostic.ai/pricing), 12 Nov 2026 | | [SentinelOne](/platforms/sentinelone/) | documented | MCP discovery plus a gateway with per-server risk scores. | [sentinelone.com](https://www.sentinelone.com/blog/prompt-security-for-agentic-ai/), 12 Nov 2026 | | [WitnessAI](/platforms/witnessai/) | documented | Catalogs MCP servers and the tools agents call. | [witness.ai](https://witness.ai/), 12 Nov 2026 | | [Koi](/platforms/koi/) | documented | Discovers MCP servers on endpoints. | [paloaltonetworks.com](https://www.paloaltonetworks.com/resources/datasheets/cortex-agentic-endpoint-security-at-a-glance), 12 Nov 2026 | | [Lasso Security](/platforms/lasso-security/) | documented | MCP discovery, risk scoring and block at connection layer; open-source MCP gateway. | [lasso.security](https://www.lasso.security/use-cases/mcp-security), 12 Nov 2026 | | [Pillar Security](/platforms/pillar-security/) | documented | Fleet MCP inventory, 23,000-server catalog risk score and per-server tool allow-lists. | [pillar.security](https://www.pillar.security/platform/agentic-endpoint), 12 Nov 2026 | | [LayerX](/platforms/layerx/) | none found | None found on 12 Nov 2026: local MCP server inventory or control. | none found on 12 Nov 2026 | | [Palo Alto Networks](/platforms/palo-alto-networks/) | none found | None found on 12 Nov 2026 for Prisma AIRS or Prisma Browser; Cortex Agentic Endpoint Security (Koi) coverage is scored on the Koi profile. | none found on 12 Nov 2026 | Ordered by documentation level, then by the site ranking. Read 13 Nov 2026. Related: [Agent Surface Map](/agent-surface-map/), [how a cell is read](/agent-surface-map/methodology/). **About Bay.** Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). ## The other agent surfaces - [Desktop apps](/surfaces/desktop-apps/) - [IDE assistants](/surfaces/ide-assistants/) - [Command-line agents](/surfaces/cli-agents/) - [Browser extensions](/surfaces/browser-extensions/) - [SaaS copilots](/surfaces/saas-copilots/) ## Sources - learn.microsoft.com/en-us/defender-endpoint/local-agent-discovery-overview: https://learn.microsoft.com/en-us/defender-endpoint/local-agent-discovery-overview (read 12 Nov 2026) - learn.microsoft.com/en-us/defender-endpoint/ai-agent-support-matrix: https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-support-matrix (read 12 Nov 2026) - www.bay.io/platform/ai-endpoint-security: https://www.bay.io/platform/ai-endpoint-security (read 12 Nov 2026) - www.bay.io/platform: https://www.bay.io/platform (read 12 Nov 2026) - www.harmonic.security/pricing: https://www.harmonic.security/pricing (read 12 Nov 2026) - docs.harmonicsecurity.app/portal-guides/managing-your-organization/configure-mcp-server-access-controls: https://docs.harmonicsecurity.app/portal-guides/managing-your-organization/configure-mcp-server-access-controls (read 12 Nov 2026) - docs.harmonicsecurity.app/portal-guides/detection-and-response/configure-scenarios-and-interventions: https://docs.harmonicsecurity.app/portal-guides/detection-and-response/configure-scenarios-and-interventions (read 12 Nov 2026) - www.catonetworks.com/platform/cato-ai-endpoint: https://www.catonetworks.com/platform/cato-ai-endpoint/ (read 12 Nov 2026) - www.crowdstrike.com/en-us/platform/falcon-guardian-aidr: https://www.crowdstrike.com/en-us/platform/falcon-guardian-aidr/ (read 12 Nov 2026) - www.crowdstrike.com/en-us/blog/crowdstrike-advances-ai-security-and-safety-falcon-guardian: https://www.crowdstrike.com/en-us/blog/crowdstrike-advances-ai-security-and-safety-falcon-guardian/ (read 12 Nov 2026) - noma.security/solutions/mcp-security: https://noma.security/solutions/mcp-security (read 12 Nov 2026) - noma.security/blog/noma-brings-agent-security-to-every-employee-endpoint: https://noma.security/blog/noma-brings-agent-security-to-every-employee-endpoint (read 12 Nov 2026) - www.island.io/product/ai: https://www.island.io/product/ai (read 12 Nov 2026) - www.island.io/blog/see-score-and-control-every-ai-asset-on-your-endpoints: https://www.island.io/blog/see-score-and-control-every-ai-asset-on-your-endpoints (read 12 Nov 2026) - zenity.io/blog/coding-agent-attack-surface: https://zenity.io/blog/coding-agent-attack-surface (read 12 Nov 2026) - zenity.io/platform/mcp-security: https://zenity.io/platform/mcp-security (read 12 Nov 2026) - www.knostic.ai/pricing: https://www.knostic.ai/pricing (read 12 Nov 2026) - www.knostic.ai/faq: https://www.knostic.ai/faq (read 12 Nov 2026) - www.knostic.ai/kirin: https://www.knostic.ai/kirin (read 12 Nov 2026) - www.sentinelone.com/blog/prompt-security-for-agentic-ai: https://www.sentinelone.com/blog/prompt-security-for-agentic-ai/ (read 12 Nov 2026) - www.sentinelone.com/blog/a-new-chapter-for-ai-and-cybersecurity-sentinelone-acquires-prompt-security: https://www.sentinelone.com/blog/a-new-chapter-for-ai-and-cybersecurity-sentinelone-acquires-prompt-security/ (read 12 Nov 2026) - www.sentinelone.com/press/sentinelone-launches-singularity-agentic-detection-and-response-to-stop-rogue-and-malicious-ai-agents-at-machine-speed: https://www.sentinelone.com/press/sentinelone-launches-singularity-agentic-detection-and-response-to-stop-rogue-and-malicious-ai-agents-at-machine-speed/ (read 12 Nov 2026) - siliconangle.com/2026/03/23/sentinelone-snyk-introduce-new-tools-securing-ai-agents: https://siliconangle.com/2026/03/23/sentinelone-snyk-introduce-new-tools-securing-ai-agents/ (read 12 Nov 2026) - witness.ai: https://witness.ai/ (read 12 Nov 2026) - witness.ai/for-employees: https://witness.ai/for-employees/ (read 12 Nov 2026) - www.paloaltonetworks.com/resources/datasheets/cortex-agentic-endpoint-security-at-a-glance: https://www.paloaltonetworks.com/resources/datasheets/cortex-agentic-endpoint-security-at-a-glance (read 12 Nov 2026) - www.paloaltonetworks.com/resources/techbriefs/koi-agentic-endpoint-security: https://www.paloaltonetworks.com/resources/techbriefs/koi-agentic-endpoint-security (read 12 Nov 2026) - www.lasso.security/use-cases/mcp-security: https://www.lasso.security/use-cases/mcp-security (read 12 Nov 2026) - github.com/lasso-security/mcp-gateway: https://github.com/lasso-security/mcp-gateway (read 12 Nov 2026) - www.pillar.security/platform/agentic-endpoint: https://www.pillar.security/platform/agentic-endpoint (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/surfaces/mcp-servers/ --- # 8 of 16 tools document coverage of browser extensions AGENT SURFACE, read 13 Nov 2026 **The short answer.** 8 of 16 AI endpoint security tools document full coverage of browser extensions, 5 document it in part, and nothing was found for 3 on 13 Nov 2026. Notes and sources for all 16 tools sit in the table below. Rows are ordered by documentation level, then by the site ranking. ## Which tools see browser extensions? 8 of 16 tools document full coverage of browser extensions: Bay, Harmonic Security, Aim Security, Island, SentinelOne, Koi, LayerX and Palo Alto Networks. 5 more do in part (Microsoft, CrowdStrike, Zenity, Knostic and WitnessAI). ## What this surface is AI extensions and agentic browsers that act on web pages, read form data and call model services from the browser. ## Tool by tool | Tool | Level | Note | Source | | --- | --- | --- | --- | | [Bay](/platforms/bay/) | documented | Bay finds browser AI tools, extensions and embedded SaaS agents and can block domains and extensions. | [bay.io](https://www.bay.io/platform/browser?utm_source=aiendpointsecurityranked.com&utm_medium=referral&utm_campaign=surface), 12 Nov 2026 | | [Harmonic Security](/platforms/harmonic-security/) | documented | Browser extension deployed alongside the Endpoint Agent; scope of AI browser agents not documented. | [harmonic.security](https://www.harmonic.security/products/command), 12 Nov 2026 | | [Aim Security](/platforms/aim-security/) | documented | Browser AI is covered by the same client and inline inspection; a Cato browser extension is named by a customer. | [catonetworks.com](https://www.catonetworks.com/platform/cato-ai-endpoint/), 12 Nov 2026 | | [Island](/platforms/island/) | documented | Own browser plus extension for Chrome, Edge, Firefox and Safari, and a browser for agents with per-agent policy. | [island.io](https://www.island.io/product/enterprise-browser-extension), 12 Nov 2026 | | [SentinelOne](/platforms/sentinelone/) | documented | Browser extension for web AI and shadow AI discovery. | [sentinelone.com](https://www.sentinelone.com/blog/a-new-chapter-for-ai-and-cybersecurity-sentinelone-acquires-prompt-security/), 12 Nov 2026 | | [Koi](/platforms/koi/) | documented | Discovers browser extensions on endpoints. | [paloaltonetworks.com](https://www.paloaltonetworks.com/resources/datasheets/cortex-agentic-endpoint-security-at-a-glance), 12 Nov 2026 | | [LayerX](/platforms/layerx/) | documented | Monitors AI browser agents and scores and blocks risky extensions. | [akamai.com](https://www.akamai.com/products/akamai-workforce-protector), 12 Nov 2026 | | [Palo Alto Networks](/platforms/palo-alto-networks/) | documented | Prisma Browser inventories and blocks risky extensions and sees GenAI and agentic workflows. | [paloaltonetworks.com](https://www.paloaltonetworks.com/sase/prisma-browser), 12 Nov 2026 | | [Microsoft](/platforms/microsoft/) | partial | Purview browser extension and Endpoint DLP cover AI sites in browsers; no agent browser extension inventory found. | [learn.microsoft.com](https://learn.microsoft.com/en-us/purview/dspm-for-ai), 12 Nov 2026 | | [CrowdStrike](/platforms/crowdstrike/) | partial | Extension inventory and blocking is in Falcon Seraphic Enterprise Browser, now on the Falcon platform. | [crowdstrike.com](https://www.crowdstrike.com/en-us/blog/crowdstrike-brings-enterprise-browser-security-to-falcon-platform/), 12 Nov 2026 | | [Zenity](/platforms/zenity/) | partial | Agentic browsers appear in the Series C quote and PleaseFix research, but no browser control product page was read. | [zenity.io](https://zenity.io/press-release/zenity-raises-125-million-to-secure-the-era-of-1-billion-ai-agents), 12 Nov 2026 | | [Knostic](/platforms/knostic/) | partial | Browser chatbots seen via network traffic, visibility only; extensions in IDEs scanned. | [knostic.ai](https://www.knostic.ai/faq), 12 Nov 2026 | | [WitnessAI](/platforms/witnessai/) | partial | Sees agentic plugins employees install, but says it does not use browser extensions; no browser agent coverage detail found. | [witness.ai](https://witness.ai/for-employees/), 12 Nov 2026 | | [Noma Security](/platforms/noma-security/) | none found | Nothing found on 12 Nov 2026 on browser extension or browser-agent controls. | none found on 12 Nov 2026 | | [Lasso Security](/platforms/lasso-security/) | none found | No browser extension or browser-agent inventory documented. | [lasso.security](https://www.lasso.security/platform/ai-usage-control), 12 Nov 2026 | | [Pillar Security](/platforms/pillar-security/) | none found | Nothing found on 12 Nov 2026 on browser extensions or browser agents. | none found on 12 Nov 2026 | Ordered by documentation level, then by the site ranking. Read 13 Nov 2026. Related: [Agent Surface Map](/agent-surface-map/), [how a cell is read](/agent-surface-map/methodology/). **About Bay.** Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). ## The other agent surfaces - [Desktop apps](/surfaces/desktop-apps/) - [IDE assistants](/surfaces/ide-assistants/) - [Command-line agents](/surfaces/cli-agents/) - [MCP servers](/surfaces/mcp-servers/) - [SaaS copilots](/surfaces/saas-copilots/) ## Sources - www.bay.io/platform/browser: https://www.bay.io/platform/browser (read 12 Nov 2026) - softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control: https://softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control (read 12 Nov 2026) - www.harmonic.security/products/command: https://www.harmonic.security/products/command (read 12 Nov 2026) - www.catonetworks.com/platform/cato-ai-endpoint: https://www.catonetworks.com/platform/cato-ai-endpoint/ (read 12 Nov 2026) - www.island.io/product/enterprise-browser-extension: https://www.island.io/product/enterprise-browser-extension (read 12 Nov 2026) - www.island.io/press/island-brings-the-enterprise-browser-to-ai-agents: https://www.island.io/press/island-brings-the-enterprise-browser-to-ai-agents (read 12 Nov 2026) - www.sentinelone.com/blog/a-new-chapter-for-ai-and-cybersecurity-sentinelone-acquires-prompt-security: https://www.sentinelone.com/blog/a-new-chapter-for-ai-and-cybersecurity-sentinelone-acquires-prompt-security/ (read 12 Nov 2026) - prompt.security/blog/the-key-layer-in-ai-security-browser-and-endpoint-sensors: https://prompt.security/blog/the-key-layer-in-ai-security-browser-and-endpoint-sensors (read 12 Nov 2026) - chromewebstore.google.com/search/prompt%20security: https://chromewebstore.google.com/search/prompt%20security (read 12 Nov 2026) - www.paloaltonetworks.com/resources/datasheets/cortex-agentic-endpoint-security-at-a-glance: https://www.paloaltonetworks.com/resources/datasheets/cortex-agentic-endpoint-security-at-a-glance (read 12 Nov 2026) - www.paloaltonetworks.com/resources/techbriefs/koi-agentic-endpoint-security: https://www.paloaltonetworks.com/resources/techbriefs/koi-agentic-endpoint-security (read 12 Nov 2026) - www.akamai.com/products/akamai-workforce-protector: https://www.akamai.com/products/akamai-workforce-protector (read 12 Nov 2026) - www.paloaltonetworks.com/sase/prisma-browser: https://www.paloaltonetworks.com/sase/prisma-browser (read 12 Nov 2026) - learn.microsoft.com/en-us/purview/dspm-for-ai: https://learn.microsoft.com/en-us/purview/dspm-for-ai (read 12 Nov 2026) - learn.microsoft.com/en-us/purview/endpoint-dlp-learn-about: https://learn.microsoft.com/en-us/purview/endpoint-dlp-learn-about (read 12 Nov 2026) - www.crowdstrike.com/en-us/blog/crowdstrike-brings-enterprise-browser-security-to-falcon-platform: https://www.crowdstrike.com/en-us/blog/crowdstrike-brings-enterprise-browser-security-to-falcon-platform/ (read 12 Nov 2026) - www.crowdstrike.com/en-us/blog/crowdstrike-shadow-AI-visibility-service: https://www.crowdstrike.com/en-us/blog/crowdstrike-shadow-AI-visibility-service/ (read 12 Nov 2026) - zenity.io/press-release/zenity-raises-125-million-to-secure-the-era-of-1-billion-ai-agents: https://zenity.io/press-release/zenity-raises-125-million-to-secure-the-era-of-1-billion-ai-agents (read 12 Nov 2026) - zenity.io/company: https://zenity.io/company (read 12 Nov 2026) - www.knostic.ai/faq: https://www.knostic.ai/faq (read 12 Nov 2026) - witness.ai/for-employees: https://witness.ai/for-employees/ (read 12 Nov 2026) - witness.ai/product: https://witness.ai/product/ (read 12 Nov 2026) - www.lasso.security/platform/ai-usage-control (checked, nothing found): https://www.lasso.security/platform/ai-usage-control (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/surfaces/browser-extensions/ --- # 3 of 16 tools document coverage of SaaS copilots AGENT SURFACE, read 13 Nov 2026 **The short answer.** 3 of 16 AI endpoint security tools document full coverage of SaaS copilots, 11 document it in part, and nothing was found for 2 on 13 Nov 2026. Notes and sources for all 16 tools sit in the table below. Rows are ordered by documentation level, then by the site ranking. ## Which tools see SaaS copilots? 3 of 16 tools document full coverage of SaaS copilots: Microsoft, Noma Security and Zenity. 11 more do in part (Bay, Harmonic Security, Aim Security, CrowdStrike, Island, Knostic, SentinelOne, WitnessAI, Lasso Security, LayerX and Palo Alto Networks). ## What this surface is Copilots and agents built into SaaS applications, which the endpoint only reaches through the browser or an API. ## Tool by tool | Tool | Level | Note | Source | | --- | --- | --- | --- | | [Microsoft](/platforms/microsoft/) | documented | Purview and Agent 365 cover Microsoft 365 Copilot, Copilot Studio and enterprise AI apps. | [learn.microsoft.com](https://learn.microsoft.com/en-us/purview/ai-microsoft-purview), 12 Nov 2026 | | [Noma Security](/platforms/noma-security/) | documented | Noma for SaaS Agents is a separate product page; Copilot Studio runtime guardrails posts exist. | [noma.security](https://noma.security/), 12 Nov 2026 | | [Zenity](/platforms/zenity/) | documented | Copilot Studio, Agentforce, ChatGPT Enterprise, Gemini covered; this is the core of the platform. | [zenity.io](https://zenity.io/), 12 Nov 2026 | | [Bay](/platforms/bay/) | partial | Bay says it surfaces embedded SaaS agents in the browser and covers SaaS applications; no named SaaS copilot list found. | [bay.io](https://www.bay.io/platform/browser?utm_source=aiendpointsecurityranked.com&utm_medium=referral&utm_campaign=surface), 12 Nov 2026 | | [Harmonic Security](/platforms/harmonic-security/) | partial | Claude for Microsoft 365 activity and Microsoft Copilot in Teams visibility documented; cloud MCP gateway early access. | [docs.harmonicsecurity.app](https://docs.harmonicsecurity.app/llms.txt), 12 Nov 2026 | | [Aim Security](/platforms/aim-security/) | partial | Microsoft Copilot use is named as securable through inline inspection; no SaaS-hosted agent controls documented in detail. | [catonetworks.com](https://www.catonetworks.com/news/cato-acquires-aim-security-to-extend-sase-leadership-and-secure-enterprise-ai-transformation/), 12 Nov 2026 | | [CrowdStrike](/platforms/crowdstrike/) | partial | Shadow SaaS agent discovery named for Copilot, Agentforce and ChatGPT Enterprise by a third party; Claude Cowork integration is pre-beta. | [msspalert.com](https://www.msspalert.com/news/crowdstrike-brings-ai-security-to-the-endpoint-and-the-timing-makes-sense), 12 Nov 2026 | | [Island](/platforms/island/) | partial | Claude for Microsoft 365 add-ins are read through the Compliance API; coverage of other SaaS copilots is described only in general terms. | [island.io](https://www.island.io/blog/claude-code-cowork-and-claude-for-microsoft-365-now-report-into-island), 12 Nov 2026 | | [Knostic](/platforms/knostic/) | partial | Earlier Knostic product covered Microsoft 365 Copilot; current Kirin scope is endpoint agents. | [siliconangle.com](https://siliconangle.com/2025/03/05/knostic-raises-11m-strengthen-enterprise-ai-security-need-know-access-controls/), 12 Nov 2026 | | [SentinelOne](/platforms/sentinelone/) | partial | ADR (early access) extends inventory to Agentforce, Copilot Studio, ChatGPT workspace agents and Claude managed agents. | [sentinelone.com](https://www.sentinelone.com/press/sentinelone-named-an-ai-security-platform-leader-in-latio-2026-ai-security-market-report/), 12 Nov 2026 | | [WitnessAI](/platforms/witnessai/) | partial | Names Windows 11 Copilot and Office 365; other SaaS copilots not named. | [witness.ai](https://witness.ai/product/), 12 Nov 2026 | | [Lasso Security](/platforms/lasso-security/) | partial | SaaS chatbots and enterprise copilots mentioned in discovery; Claude Enterprise managed settings. | [lasso.security](https://www.lasso.security/platform/ai-usage-control), 12 Nov 2026 | | [LayerX](/platforms/layerx/) | partial | Controls SaaS and web app use through the browser; named SaaS copilots are not listed. | [akamai.com](https://www.akamai.com/products/akamai-workforce-protector), 12 Nov 2026 | | [Palo Alto Networks](/platforms/palo-alto-networks/) | partial | Prisma Browser controls actions in SaaS and GenAI apps; specific copilots are not named. | [paloaltonetworks.com](https://www.paloaltonetworks.com/sase/prisma-browser), 12 Nov 2026 | | [Koi](/platforms/koi/) | none found | No SaaS-hosted copilot coverage found in AES material. | none found on 12 Nov 2026 | | [Pillar Security](/platforms/pillar-security/) | none found | Nothing found on 12 Nov 2026 for SaaS-hosted copilots on the endpoint page. | none found on 12 Nov 2026 | Ordered by documentation level, then by the site ranking. Read 13 Nov 2026. Related: [Agent Surface Map](/agent-surface-map/), [how a cell is read](/agent-surface-map/methodology/). **About Bay.** Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). ## The other agent surfaces - [Desktop apps](/surfaces/desktop-apps/) - [IDE assistants](/surfaces/ide-assistants/) - [Command-line agents](/surfaces/cli-agents/) - [MCP servers](/surfaces/mcp-servers/) - [Browser extensions](/surfaces/browser-extensions/) ## Sources - learn.microsoft.com/en-us/purview/ai-microsoft-purview: https://learn.microsoft.com/en-us/purview/ai-microsoft-purview (read 12 Nov 2026) - www.microsoft.com/en-us/security/blog/2026/05/01/microsoft-agent-365-now-generally-available-expands-capabilities-and-integrations: https://www.microsoft.com/en-us/security/blog/2026/05/01/microsoft-agent-365-now-generally-available-expands-capabilities-and-integrations/ (read 12 Nov 2026) - noma.security: https://noma.security/ (read 12 Nov 2026) - noma.security/platform/endpoint-agents: https://noma.security/platform/endpoint-agents (read 12 Nov 2026) - zenity.io: https://zenity.io/ (read 12 Nov 2026) - zenity.io/press-release/zenity-raises-125-million-to-secure-the-era-of-1-billion-ai-agents: https://zenity.io/press-release/zenity-raises-125-million-to-secure-the-era-of-1-billion-ai-agents (read 12 Nov 2026) - zenity.io/press-release/zenity-extends-ai-agent-security-and-governance-to-claude-enterprise: https://zenity.io/press-release/zenity-extends-ai-agent-security-and-governance-to-claude-enterprise (read 12 Nov 2026) - www.bay.io/platform/browser: https://www.bay.io/platform/browser (read 12 Nov 2026) - www.bay.io/blog/bay-emerges-from-stealth-with-50m-after-fortune-100-companies-used-its-platform-to-triple-autonomous-ai-agent-use-2: https://www.bay.io/blog/bay-emerges-from-stealth-with-50m-after-fortune-100-companies-used-its-platform-to-triple-autonomous-ai-agent-use-2 (read 12 Nov 2026) - docs.harmonicsecurity.app/llms.txt: https://docs.harmonicsecurity.app/llms.txt (read 12 Nov 2026) - docs.harmonicsecurity.app/mcp-gateway/getting-started/what-is-mcp-gateway: https://docs.harmonicsecurity.app/mcp-gateway/getting-started/what-is-mcp-gateway (read 12 Nov 2026) - docs.harmonicsecurity.app/portal-guides/integration-guides/configure-anthropic-compliance-api-connector: https://docs.harmonicsecurity.app/portal-guides/integration-guides/configure-anthropic-compliance-api-connector (read 12 Nov 2026) - www.catonetworks.com/news/cato-acquires-aim-security-to-extend-sase-leadership-and-secure-enterprise-ai-transformation: https://www.catonetworks.com/news/cato-acquires-aim-security-to-extend-sase-leadership-and-secure-enterprise-ai-transformation/ (read 12 Nov 2026) - www.msspalert.com/news/crowdstrike-brings-ai-security-to-the-endpoint-and-the-timing-makes-sense: https://www.msspalert.com/news/crowdstrike-brings-ai-security-to-the-endpoint-and-the-timing-makes-sense (read 12 Nov 2026) - www.crowdstrike.com/en-us/blog/crowdstrike-advances-ai-security-and-safety-falcon-guardian: https://www.crowdstrike.com/en-us/blog/crowdstrike-advances-ai-security-and-safety-falcon-guardian/ (read 12 Nov 2026) - www.island.io/blog/claude-code-cowork-and-claude-for-microsoft-365-now-report-into-island: https://www.island.io/blog/claude-code-cowork-and-claude-for-microsoft-365-now-report-into-island (read 12 Nov 2026) - siliconangle.com/2025/03/05/knostic-raises-11m-strengthen-enterprise-ai-security-need-know-access-controls: https://siliconangle.com/2025/03/05/knostic-raises-11m-strengthen-enterprise-ai-security-need-know-access-controls/ (read 12 Nov 2026) - www.sentinelone.com/press/sentinelone-named-an-ai-security-platform-leader-in-latio-2026-ai-security-market-report: https://www.sentinelone.com/press/sentinelone-named-an-ai-security-platform-leader-in-latio-2026-ai-security-market-report/ (read 12 Nov 2026) - www.sentinelone.com/press/sentinelone-launches-singularity-agentic-detection-and-response-to-stop-rogue-and-malicious-ai-agents-at-machine-speed: https://www.sentinelone.com/press/sentinelone-launches-singularity-agentic-detection-and-response-to-stop-rogue-and-malicious-ai-agents-at-machine-speed/ (read 12 Nov 2026) - witness.ai/product: https://witness.ai/product/ (read 12 Nov 2026) - www.lasso.security/platform/ai-usage-control: https://www.lasso.security/platform/ai-usage-control (read 12 Nov 2026) - www.lasso.security: https://www.lasso.security (read 12 Nov 2026) - www.lasso.security/blog/bringing-claude-enterprise-activity-into-lasso-with-claudes-compliance-api: https://www.lasso.security/blog/bringing-claude-enterprise-activity-into-lasso-with-claudes-compliance-api (read 12 Nov 2026) - www.akamai.com/products/akamai-workforce-protector: https://www.akamai.com/products/akamai-workforce-protector (read 12 Nov 2026) - www.paloaltonetworks.com/sase/prisma-browser: https://www.paloaltonetworks.com/sase/prisma-browser (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/surfaces/saas-copilots/ --- # Six places agents run on an endpoint, and how many tools cover each AGENT SURFACES, read 13 Nov 2026 **The short answer.** The Agent Surface Map splits into 6 surfaces: desktop apps (9 of 16), IDE assistants (14 of 16), command-line agents (12 of 16), MCP servers (14 of 16), browser extensions (8 of 16), SaaS copilots (3 of 16). The counts are tools with a documented capability on 13 Nov 2026. - [Desktop apps](/surfaces/desktop-apps/): 9 of 16 tools documented. Chat and agent applications installed on a laptop, such as a desktop assistant that can read local folders or run tools on the user's behalf. - [IDE assistants](/surfaces/ide-assistants/): 14 of 16 tools documented. Assistants inside code editors and IDEs, which read the open repository and can propose or apply edits and commands. - [Command-line agents](/surfaces/cli-agents/): 12 of 16 tools documented. Coding agents that run in a terminal with the user's shell, files and credentials, usually started by a developer. - [MCP servers](/surfaces/mcp-servers/): 14 of 16 tools documented. Model Context Protocol servers running on or reached from the endpoint, which hand agents tools, files and data sources. - [Browser extensions](/surfaces/browser-extensions/): 8 of 16 tools documented. AI extensions and agentic browsers that act on web pages, read form data and call model services from the browser. - [SaaS copilots](/surfaces/saas-copilots/): 3 of 16 tools documented. Copilots and agents built into SaaS applications, which the endpoint only reaches through the browser or an API. ## Sources - Agent Surface Map: https://aiendpointsecurityranked.com/agent-surface-map/ (read 13 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/surfaces/ --- # 4 of 16 tools document stopping file read before it completes AGENT ACTION, read 13 Nov 2026 **The short answer.** 4 of 16 AI endpoint security tools document stopping file read, 5 document it in part, and nothing was found for 7 on 13 Nov 2026. Notes and sources for all 16 tools sit in the table below. Rows are ordered by documentation level, then by the site ranking. ## Which tools can stop a file read? 4 of 16 tools document stopping file read before it completes: Noma Security, Island, Knostic and Pillar Security. 5 more do in part (Microsoft, Bay, Aim Security, CrowdStrike and Zenity). ## What this action is An agent opening a local file, including files outside the project it was started in. ## Tool by tool | Tool | Level | Note | Source | | --- | --- | --- | --- | | [Noma Security](/platforms/noma-security/) | documented | Noma's public Claude Code plugin README documents hook-based evaluation of file reads (.env, SSH keys) with block, mask and ask policies (beta; fail-open on hook failure). | [noma.security](https://noma.security/blog/noma-brings-agent-security-to-every-employee-endpoint), 12 Nov 2026 | | [Island](/platforms/island/) | documented | Says the file read on the endpoint can be blocked in an agent chain. | [island.io](https://www.island.io/blog/claude-code-cowork-and-claude-for-microsoft-365-now-report-into-island), 12 Nov 2026 | | [Knostic](/platforms/knostic/) | documented | Enforce mode blocks reads of files containing secrets. | [knostic.ai](https://www.knostic.ai/faq), 12 Nov 2026 | | [Pillar Security](/platforms/pillar-security/) | documented | Blocks reads of keys.env, SSH and cloud credentials from reaching external destinations. | [pillar.security](https://www.pillar.security/platform/agentic-endpoint), 12 Nov 2026 | | [Microsoft](/platforms/microsoft/) | partial | Example blocks reading a local environment file after injection; general not file-specific. | [learn.microsoft.com](https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-runtime-protection-overview), 12 Nov 2026 | | [Bay](/platforms/bay/) | partial | Analyst profile lists files among evaluated action types; Bay pages do not detail file-read blocking. | [softwareanalyst.substack.com](https://softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control), 12 Nov 2026 | | [Aim Security](/platforms/aim-security/) | partial | Dashboard flags risky settings such as writes outside the project; no documented block of file reads. | [catonetworks.com](https://www.catonetworks.com/platform/cato-ai-endpoint/), 12 Nov 2026 | | [CrowdStrike](/platforms/crowdstrike/) | partial | Sensor sees files agents access; blocking is documented via agent hooks in general terms. | [crowdstrike.com](https://www.crowdstrike.com/en-us/platform/falcon-guardian-aidr/), 12 Nov 2026 | | [Zenity](/platforms/zenity/) | partial | Sensitive-file access followed by network action is blocked; plain file reads are monitored. | [zenity.io](https://zenity.io/blog/coding-agent-attack-surface), 12 Nov 2026 | | [Harmonic Security](/platforms/harmonic-security/) | none found | Documented controls cover prompts and tool calls; file reads not named. | [docs.harmonicsecurity.app](https://docs.harmonicsecurity.app/portal-guides/detection-and-response/sensitive-data-controls-for-coding-agents-and-desktop-applications), 12 Nov 2026 | | [SentinelOne](/platforms/sentinelone/) | none found | None found on 12 Nov 2026. | none found on 12 Nov 2026 | | [WitnessAI](/platforms/witnessai/) | none found | No file-read blocking documented. | none found on 12 Nov 2026 | | [Koi](/platforms/koi/) | none found | No file-read blocking documented. | [paloaltonetworks.com](https://www.paloaltonetworks.com/cortex/agentic-endpoint-security), 12 Nov 2026 | | [Lasso Security](/platforms/lasso-security/) | none found | File reads not named as a blockable action. | [lasso.security](https://www.lasso.security/use-cases/ai-coding-assistants), 12 Nov 2026 | | [LayerX](/platforms/layerx/) | none found | None found on 12 Nov 2026. | none found on 12 Nov 2026 | | [Palo Alto Networks](/platforms/palo-alto-networks/) | none found | No file-read blocking documented. | [paloaltonetworks.com](https://www.paloaltonetworks.com/prisma/prisma-ai-runtime-security), 12 Nov 2026 | Ordered by documentation level, then by the site ranking. Read 13 Nov 2026. Related: [Action Control Matrix](/action-control-matrix/), [how a cell is read](/action-control-matrix/methodology/). **About Bay.** Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). ## The other agent actions - [Shell command](/actions/shell-command/) - [Network call](/actions/network-call/) - [Credential access](/actions/credential-access/) - [Data paste](/actions/data-paste/) - [Package install](/actions/package-install/) ## Sources - noma.security/blog/noma-brings-agent-security-to-every-employee-endpoint: https://noma.security/blog/noma-brings-agent-security-to-every-employee-endpoint (read 12 Nov 2026) - github.com/Noma-Security/noma-marketplace/blob/main/guardrails/README.md: https://github.com/Noma-Security/noma-marketplace/blob/main/guardrails/README.md (read 12 Nov 2026) - www.island.io/blog/claude-code-cowork-and-claude-for-microsoft-365-now-report-into-island: https://www.island.io/blog/claude-code-cowork-and-claude-for-microsoft-365-now-report-into-island (read 12 Nov 2026) - www.knostic.ai/faq: https://www.knostic.ai/faq (read 12 Nov 2026) - www.pillar.security/platform/agentic-endpoint: https://www.pillar.security/platform/agentic-endpoint (read 12 Nov 2026) - learn.microsoft.com/en-us/defender-endpoint/ai-agent-runtime-protection-overview: https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-runtime-protection-overview (read 12 Nov 2026) - softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control: https://softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control (read 12 Nov 2026) - www.catonetworks.com/platform/cato-ai-endpoint: https://www.catonetworks.com/platform/cato-ai-endpoint/ (read 12 Nov 2026) - www.crowdstrike.com/en-us/platform/falcon-guardian-aidr: https://www.crowdstrike.com/en-us/platform/falcon-guardian-aidr/ (read 12 Nov 2026) - www.crowdstrike.com/en-us/blog/crowdstrike-advances-ai-security-and-safety-falcon-guardian: https://www.crowdstrike.com/en-us/blog/crowdstrike-advances-ai-security-and-safety-falcon-guardian/ (read 12 Nov 2026) - zenity.io/blog/coding-agent-attack-surface: https://zenity.io/blog/coding-agent-attack-surface (read 12 Nov 2026) - docs.harmonicsecurity.app/portal-guides/detection-and-response/sensitive-data-controls-for-coding-agents-and-desktop-applications (checked, nothing found): https://docs.harmonicsecurity.app/portal-guides/detection-and-response/sensitive-data-controls-for-coding-agents-and-desktop-applications (read 12 Nov 2026) - www.paloaltonetworks.com/cortex/agentic-endpoint-security (checked, nothing found): https://www.paloaltonetworks.com/cortex/agentic-endpoint-security (read 12 Nov 2026) - www.lasso.security/use-cases/ai-coding-assistants (checked, nothing found): https://www.lasso.security/use-cases/ai-coding-assistants (read 12 Nov 2026) - www.paloaltonetworks.com/prisma/prisma-ai-runtime-security (checked, nothing found): https://www.paloaltonetworks.com/prisma/prisma-ai-runtime-security (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/actions/file-read/ --- # 7 of 16 tools document stopping shell command before it completes AGENT ACTION, read 13 Nov 2026 **The short answer.** 7 of 16 AI endpoint security tools document stopping shell command, 8 document it in part, and nothing was found for 1 on 13 Nov 2026. Notes and sources for all 16 tools sit in the table below. Rows are ordered by documentation level, then by the site ranking. ## Which tools can stop a shell command before it runs? 7 of 16 tools document stopping shell command before it completes: Microsoft, Bay, Noma Security, Island, Zenity, Knostic and Pillar Security. 8 more do in part (Harmonic Security, Aim Security, CrowdStrike, SentinelOne, WitnessAI, Koi, Lasso Security and Palo Alto Networks). ## What this action is An agent running a command in a shell on the endpoint. ## Tool by tool | Tool | Level | Note | Source | | --- | --- | --- | --- | | [Microsoft](/platforms/microsoft/) | documented | Agent-native inspection can prevent a harmful tool action from running in supported agents (preview, Windows). | [learn.microsoft.com](https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-runtime-protection-overview), 12 Nov 2026 | | [Bay](/platforms/bay/) | documented | Bay names risky shell commands as an action it allows, asks, denies or steers. | [bay.io](https://www.bay.io/platform/ai-endpoint-security?utm_source=aiendpointsecurityranked.com&utm_medium=referral&utm_campaign=action), 12 Nov 2026 | | [Noma Security](/platforms/noma-security/) | documented | Noma's public Claude Code plugin README documents hook-based evaluation of shell commands with block, mask and ask policies (beta; fail-open on hook failure). | [noma.security](https://noma.security/blog/securing-the-agentic-frontier-noma-unveils-the-first-real-time-agent-runtime-security-for-cursor), 12 Nov 2026 | | [Island](/platforms/island/) | documented | Bash execution can be blocked, or the user can be prompted, per policy. | [island.io](https://www.island.io/product/ai), 12 Nov 2026 | | [Zenity](/platforms/zenity/) | documented | Block or modify a dangerous tool call (commands) before execution via hooks. | [zenity.io](https://zenity.io/use-cases/agent-type/coding-personal-agents), 12 Nov 2026 | | [Knostic](/platforms/knostic/) | documented | Demo shows destructive command blocked before it ran; Kirin page claims block of risky commands. | [knostic.ai](https://www.knostic.ai), 12 Nov 2026 | | [Pillar Security](/platforms/pillar-security/) | documented | Pre-tool hook decisions before each action; risky commands blocked before execution per the page. | [pillar.security](https://www.pillar.security/platform/agentic-endpoint), 12 Nov 2026 | | [Harmonic Security](/platforms/harmonic-security/) | partial | Tool calls are intercepted through hooks, shell commands not named specifically. | [docs.harmonicsecurity.app](https://docs.harmonicsecurity.app/portal-guides/detection-and-response/sensitive-data-controls-for-coding-agents-and-desktop-applications), 12 Nov 2026 | | [Aim Security](/platforms/aim-security/) | partial | Flags full command-line access; blocks unapproved tool calls in supported agents, not shell commands specifically. | [catonetworks.com](https://www.catonetworks.com/platform/cato-ai-endpoint/), 12 Nov 2026 | | [CrowdStrike](/platforms/crowdstrike/) | partial | Sensor sees commands agents run; hooks can block malicious activity before execution in supported agents. | [crowdstrike.com](https://www.crowdstrike.com/en-us/platform/falcon-guardian-aidr/), 12 Nov 2026 | | [SentinelOne](/platforms/sentinelone/) | partial | Tool-call blocking and a Cowork VM behavioral engine; shell commands not named. | [sentinelone.com](https://www.sentinelone.com/blog/prompt-security-for-agentic-ai/), 12 Nov 2026 | | [WitnessAI](/platforms/witnessai/) | partial | Captures shell executions for supported agents; blocking of shell commands is not documented. | [witness.ai](https://witness.ai/for-agents/), 12 Nov 2026 | | [Koi](/platforms/koi/) | partial | Blocks rogue AI agent actions in general terms; shell commands not named. | [paloaltonetworks.com](https://www.paloaltonetworks.com/resources/datasheets/cortex-agentic-endpoint-security-at-a-glance), 12 Nov 2026 | | [Lasso Security](/platforms/lasso-security/) | partial | Tool calls inspected before execution and flagged or blocked; shell not named as a distinct control. | [lasso.security](https://www.lasso.security/use-cases/ai-coding-assistants), 12 Nov 2026 | | [Palo Alto Networks](/platforms/palo-alto-networks/) | partial | Cortex XDR One guardrails are said to help stop destructive commands (Oct 2026); shell commands are not itemized and AES docs were not found. | [paloaltonetworks.com](https://www.paloaltonetworks.com/prisma/prisma-ai-runtime-security), 12 Nov 2026 | | [LayerX](/platforms/layerx/) | none found | None found on 12 Nov 2026. | none found on 12 Nov 2026 | Ordered by documentation level, then by the site ranking. Read 13 Nov 2026. Related: [Action Control Matrix](/action-control-matrix/), [how a cell is read](/action-control-matrix/methodology/). **About Bay.** Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). ## The other agent actions - [File read](/actions/file-read/) - [Network call](/actions/network-call/) - [Credential access](/actions/credential-access/) - [Data paste](/actions/data-paste/) - [Package install](/actions/package-install/) ## Sources - learn.microsoft.com/en-us/defender-endpoint/ai-agent-runtime-protection-overview: https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-runtime-protection-overview (read 12 Nov 2026) - www.bay.io/platform/ai-endpoint-security: https://www.bay.io/platform/ai-endpoint-security (read 12 Nov 2026) - noma.security/blog/securing-the-agentic-frontier-noma-unveils-the-first-real-time-agent-runtime-security-for-cursor: https://noma.security/blog/securing-the-agentic-frontier-noma-unveils-the-first-real-time-agent-runtime-security-for-cursor (read 12 Nov 2026) - noma.security/platform/endpoint-agents: https://noma.security/platform/endpoint-agents (read 12 Nov 2026) - github.com/Noma-Security/noma-marketplace/blob/main/guardrails/README.md: https://github.com/Noma-Security/noma-marketplace/blob/main/guardrails/README.md (read 12 Nov 2026) - www.island.io/product/ai: https://www.island.io/product/ai (read 12 Nov 2026) - www.island.io/blog/ai-runtime-protection-governing-what-agents-do-not-just-what-theyre-told: https://www.island.io/blog/ai-runtime-protection-governing-what-agents-do-not-just-what-theyre-told (read 12 Nov 2026) - zenity.io/use-cases/agent-type/coding-personal-agents: https://zenity.io/use-cases/agent-type/coding-personal-agents (read 12 Nov 2026) - zenity.io/platform/ai-detection-and-response: https://zenity.io/platform/ai-detection-and-response (read 12 Nov 2026) - www.knostic.ai: https://www.knostic.ai (read 12 Nov 2026) - www.knostic.ai/kirin: https://www.knostic.ai/kirin (read 12 Nov 2026) - www.pillar.security/platform/agentic-endpoint: https://www.pillar.security/platform/agentic-endpoint (read 12 Nov 2026) - docs.harmonicsecurity.app/portal-guides/detection-and-response/sensitive-data-controls-for-coding-agents-and-desktop-applications: https://docs.harmonicsecurity.app/portal-guides/detection-and-response/sensitive-data-controls-for-coding-agents-and-desktop-applications (read 12 Nov 2026) - www.catonetworks.com/platform/cato-ai-endpoint: https://www.catonetworks.com/platform/cato-ai-endpoint/ (read 12 Nov 2026) - www.catonetworks.com/platform/agentic-ai-security: https://www.catonetworks.com/platform/agentic-ai-security/ (read 12 Nov 2026) - www.crowdstrike.com/en-us/platform/falcon-guardian-aidr: https://www.crowdstrike.com/en-us/platform/falcon-guardian-aidr/ (read 12 Nov 2026) - www.crowdstrike.com/en-us/blog/crowdstrike-advances-ai-security-and-safety-falcon-guardian: https://www.crowdstrike.com/en-us/blog/crowdstrike-advances-ai-security-and-safety-falcon-guardian/ (read 12 Nov 2026) - www.sentinelone.com/blog/prompt-security-for-agentic-ai: https://www.sentinelone.com/blog/prompt-security-for-agentic-ai/ (read 12 Nov 2026) - www.sentinelone.com/press/sentinelone-expands-ai-security-coverage-to-claude-cowork: https://www.sentinelone.com/press/sentinelone-expands-ai-security-coverage-to-claude-cowork/ (read 12 Nov 2026) - www.sentinelone.com/press/sentinelone-launches-singularity-agentic-detection-and-response-to-stop-rogue-and-malicious-ai-agents-at-machine-speed: https://www.sentinelone.com/press/sentinelone-launches-singularity-agentic-detection-and-response-to-stop-rogue-and-malicious-ai-agents-at-machine-speed/ (read 12 Nov 2026) - witness.ai/for-agents: https://witness.ai/for-agents/ (read 12 Nov 2026) - www.paloaltonetworks.com/resources/datasheets/cortex-agentic-endpoint-security-at-a-glance: https://www.paloaltonetworks.com/resources/datasheets/cortex-agentic-endpoint-security-at-a-glance (read 12 Nov 2026) - www.paloaltonetworks.com/cortex/agentic-endpoint-security (checked, nothing found): https://www.paloaltonetworks.com/cortex/agentic-endpoint-security (read 12 Nov 2026) - www.lasso.security/use-cases/ai-coding-assistants: https://www.lasso.security/use-cases/ai-coding-assistants (read 12 Nov 2026) - www.paloaltonetworks.com/prisma/prisma-ai-runtime-security: https://www.paloaltonetworks.com/prisma/prisma-ai-runtime-security (read 12 Nov 2026) - www.paloaltonetworks.com/blog/2026/10/reinventing-endpoint-security-with-cortex-xdr-one: https://www.paloaltonetworks.com/blog/2026/10/reinventing-endpoint-security-with-cortex-xdr-one/ (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/actions/shell-command/ --- # 5 of 16 tools document stopping network call before it completes AGENT ACTION, read 13 Nov 2026 **The short answer.** 5 of 16 AI endpoint security tools document stopping network call, 10 document it in part, and nothing was found for 1 on 13 Nov 2026. Notes and sources for all 16 tools sit in the table below. Rows are ordered by documentation level, then by the site ranking. ## Which tools can stop a network call? 5 of 16 tools document stopping network call before it completes: Bay, Island, Zenity, WitnessAI and Pillar Security. 10 more do in part (Microsoft, Harmonic Security, Aim Security, CrowdStrike, Noma Security, Knostic, SentinelOne, Koi, Lasso Security and Palo Alto Networks). ## What this action is An agent sending a request to an outside host or service. ## Tool by tool | Tool | Level | Note | Source | | --- | --- | --- | --- | | [Bay](/platforms/bay/) | documented | Analyst profile lists network destinations and exfiltration-route closure, as stated by Bay. | [softwareanalyst.substack.com](https://softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control), 12 Nov 2026 | | [Island](/platforms/island/) | documented | Says the outbound connection can be blocked on the network; malicious URLs are also blocked. | [island.io](https://www.island.io/blog/claude-code-cowork-and-claude-for-microsoft-365-now-report-into-island), 12 Nov 2026 | | [Zenity](/platforms/zenity/) | documented | Network-bound action after sensitive access is blocked at PreToolUse. | [zenity.io](https://zenity.io/blog/coding-agent-attack-surface), 12 Nov 2026 | | [WitnessAI](/platforms/witnessai/) | documented | Blocks agent tool calls and MCP server connections inline in the traffic path. | [witness.ai](https://witness.ai/product/), 12 Nov 2026 | | [Pillar Security](/platforms/pillar-security/) | documented | Credential-read flows to external destinations blocked. | [pillar.security](https://www.pillar.security/platform/agentic-endpoint), 12 Nov 2026 | | [Microsoft](/platforms/microsoft/) | partial | Blocks data leaving after injection; network inspection excludes pinned and HTTP/3 agents. | [learn.microsoft.com](https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-runtime-protection-overview), 12 Nov 2026 | | [Harmonic Security](/platforms/harmonic-security/) | partial | MCP gateway blocks unpermitted server access and tool calls; general network calls not named. | [docs.harmonicsecurity.app](https://docs.harmonicsecurity.app/portal-guides/managing-your-organization/configure-mcp-server-access-controls), 12 Nov 2026 | | [Aim Security](/platforms/aim-security/) | partial | Inline network inspection of AI traffic and tool-call blocking; no documented blocking of arbitrary agent network calls. | [catonetworks.com](https://www.catonetworks.com/platform/cato-ai-endpoint/), 12 Nov 2026 | | [CrowdStrike](/platforms/crowdstrike/) | partial | Kill switch blocks model access at a gateway; endpoint network inspection is pre-beta. | [crowdstrike.com](https://www.crowdstrike.com/en-us/platform/falcon-guardian-aidr/), 12 Nov 2026 | | [Noma Security](/platforms/noma-security/) | partial | Block and route-to-human decisions at the action level; network-specific control not itemized. | [noma.security](https://noma.security/blog/noma-brings-agent-security-to-every-employee-endpoint), 12 Nov 2026 | | [Knostic](/platforms/knostic/) | partial | Risky MCP servers quarantined so agents cannot connect. | [knostic.ai](https://www.knostic.ai/faq), 12 Nov 2026 | | [SentinelOne](/platforms/sentinelone/) | partial | MCP calls are intercepted and can be blocked; other network calls not named. | [sentinelone.com](https://www.sentinelone.com/blog/a-new-chapter-for-ai-and-cybersecurity-sentinelone-acquires-prompt-security/), 12 Nov 2026 | | [Koi](/platforms/koi/) | partial | Blocks malicious URLs via Prisma AIRS integration. | [paloaltonetworks.com](https://www.paloaltonetworks.com/cortex/agentic-endpoint-security.md), 12 Nov 2026 | | [Lasso Security](/platforms/lasso-security/) | partial | Blocks risky MCP servers and injection at the connection layer. | [lasso.security](https://www.lasso.security/use-cases/mcp-security), 12 Nov 2026 | | [Palo Alto Networks](/platforms/palo-alto-networks/) | partial | AIRS Network Intercept mode inspects AI traffic; blocking specifics not read. | [paloaltonetworks.com](https://www.paloaltonetworks.com/prisma/prisma-ai-runtime-security), 12 Nov 2026 | | [LayerX](/platforms/layerx/) | none found | None found on 12 Nov 2026: outbound call blocking outside the browser. | none found on 12 Nov 2026 | Ordered by documentation level, then by the site ranking. Read 13 Nov 2026. Related: [Action Control Matrix](/action-control-matrix/), [how a cell is read](/action-control-matrix/methodology/). **About Bay.** Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). ## The other agent actions - [File read](/actions/file-read/) - [Shell command](/actions/shell-command/) - [Credential access](/actions/credential-access/) - [Data paste](/actions/data-paste/) - [Package install](/actions/package-install/) ## Sources - softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control: https://softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control (read 12 Nov 2026) - www.island.io/blog/claude-code-cowork-and-claude-for-microsoft-365-now-report-into-island: https://www.island.io/blog/claude-code-cowork-and-claude-for-microsoft-365-now-report-into-island (read 12 Nov 2026) - zenity.io/blog/coding-agent-attack-surface: https://zenity.io/blog/coding-agent-attack-surface (read 12 Nov 2026) - witness.ai/product: https://witness.ai/product/ (read 12 Nov 2026) - witness.ai/for-agents: https://witness.ai/for-agents/ (read 12 Nov 2026) - witness.ai/blog/witnessai-in-sacrs-arise-market-map-why-a-unified-ai-control-plane-matters: https://witness.ai/blog/witnessai-in-sacrs-arise-market-map-why-a-unified-ai-control-plane-matters/ (read 12 Nov 2026) - www.pillar.security/platform/agentic-endpoint: https://www.pillar.security/platform/agentic-endpoint (read 12 Nov 2026) - learn.microsoft.com/en-us/defender-endpoint/ai-agent-runtime-protection-overview: https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-runtime-protection-overview (read 12 Nov 2026) - docs.harmonicsecurity.app/portal-guides/managing-your-organization/configure-mcp-server-access-controls: https://docs.harmonicsecurity.app/portal-guides/managing-your-organization/configure-mcp-server-access-controls (read 12 Nov 2026) - www.harmonic.security/pricing: https://www.harmonic.security/pricing (read 12 Nov 2026) - www.catonetworks.com/platform/cato-ai-endpoint: https://www.catonetworks.com/platform/cato-ai-endpoint/ (read 12 Nov 2026) - www.catonetworks.com/platform/agentic-ai-security: https://www.catonetworks.com/platform/agentic-ai-security/ (read 12 Nov 2026) - www.crowdstrike.com/en-us/platform/falcon-guardian-aidr: https://www.crowdstrike.com/en-us/platform/falcon-guardian-aidr/ (read 12 Nov 2026) - www.crowdstrike.com/en-us/blog/crowdstrike-advances-ai-security-and-safety-falcon-guardian: https://www.crowdstrike.com/en-us/blog/crowdstrike-advances-ai-security-and-safety-falcon-guardian/ (read 12 Nov 2026) - noma.security/blog/noma-brings-agent-security-to-every-employee-endpoint: https://noma.security/blog/noma-brings-agent-security-to-every-employee-endpoint (read 12 Nov 2026) - noma.security/platform/endpoint-agents: https://noma.security/platform/endpoint-agents (read 12 Nov 2026) - www.knostic.ai/faq: https://www.knostic.ai/faq (read 12 Nov 2026) - www.sentinelone.com/blog/a-new-chapter-for-ai-and-cybersecurity-sentinelone-acquires-prompt-security: https://www.sentinelone.com/blog/a-new-chapter-for-ai-and-cybersecurity-sentinelone-acquires-prompt-security/ (read 12 Nov 2026) - www.sentinelone.com/blog/prompt-security-for-agentic-ai: https://www.sentinelone.com/blog/prompt-security-for-agentic-ai/ (read 12 Nov 2026) - www.paloaltonetworks.com/cortex/agentic-endpoint-security.md: https://www.paloaltonetworks.com/cortex/agentic-endpoint-security.md (read 12 Nov 2026) - www.lasso.security/use-cases/mcp-security: https://www.lasso.security/use-cases/mcp-security (read 12 Nov 2026) - github.com/lasso-security/kong-ai-lasso-guardrail: https://github.com/lasso-security/kong-ai-lasso-guardrail (read 12 Nov 2026) - www.paloaltonetworks.com/prisma/prisma-ai-runtime-security: https://www.paloaltonetworks.com/prisma/prisma-ai-runtime-security (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/actions/network-call/ --- # 4 of 16 tools document stopping credential access before it completes AGENT ACTION, read 13 Nov 2026 **The short answer.** 4 of 16 AI endpoint security tools document stopping credential access, 8 document it in part, and nothing was found for 4 on 13 Nov 2026. Notes and sources for all 16 tools sit in the table below. Rows are ordered by documentation level, then by the site ranking. ## Which tools can stop credential access? 4 of 16 tools document stopping credential access before it completes: Bay, Zenity, Knostic and Pillar Security. 8 more do in part (Microsoft, Harmonic Security, CrowdStrike, Noma Security, Island, SentinelOne, WitnessAI and Lasso Security). ## What this action is An agent reading keys, tokens, cookies or password stores. ## Tool by tool | Tool | Level | Note | Source | | --- | --- | --- | --- | | [Bay](/platforms/bay/) | documented | Bay names credential access as an action it controls. | [bay.io](https://www.bay.io/platform/ai-endpoint-security?utm_source=aiendpointsecurityranked.com&utm_medium=referral&utm_campaign=action), 12 Nov 2026 | | [Zenity](/platforms/zenity/) | documented | Secret exfiltration prevention named.env, SSH keys, credential stores in the blog. | [zenity.io](https://zenity.io/blog/coding-agent-attack-surface), 12 Nov 2026 | | [Knostic](/platforms/knostic/) | documented | Secret reads blocked or sanitized; secrets in prompts blocked by hook. | [knostic.ai](https://www.knostic.ai/faq), 12 Nov 2026 | | [Pillar Security](/platforms/pillar-security/) | documented | Credential exfiltration blocking named explicitly. | [pillar.security](https://www.pillar.security/platform/agentic-endpoint), 12 Nov 2026 | | [Microsoft](/platforms/microsoft/) | partial | Example covers environment file contents; credential-specific blocking not documented. | [learn.microsoft.com](https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-runtime-protection-overview), 12 Nov 2026 | | [Harmonic Security](/platforms/harmonic-security/) | partial | Sensitive data detection blocks credentials in prompts; credential-file access not named. | [harmonic.security](https://www.harmonic.security/pricing), 12 Nov 2026 | | [CrowdStrike](/platforms/crowdstrike/) | partial | Credentials in prompts are blocked before reaching models; credential file access blocking not specifically documented. | [crowdstrike.com](https://www.crowdstrike.com/en-us/press-releases/crowdstrike-announces-general-availability-of-falcon-ai-detection-and-response/), 12 Nov 2026 | | [Noma Security](/platforms/noma-security/) | partial | Credential leakage named as a risk and secrets detection in AI-DR; block per detector setting. | [noma.security](https://noma.security/blog/noma-brings-agent-security-to-every-employee-endpoint), 12 Nov 2026 | | [Island](/platforms/island/) | partial | Time-boxed credentials and a browser vault are described; blocking of local credential file access is not documented. | [island.io](https://www.island.io/product/ai), 12 Nov 2026 | | [SentinelOne](/platforms/sentinelone/) | partial | ADR (early access, Oct 2026) says it blocks credential exfiltration before the action executes; endpoint file and credential access are not itemized. | [sentinelone.com](https://www.sentinelone.com/press/sentinelone-launches-singularity-agentic-detection-and-response-to-stop-rogue-and-malicious-ai-agents-at-machine-speed/), 12 Nov 2026 | | [WitnessAI](/platforms/witnessai/) | partial | Tokenizes credentials and secrets before they reach a model or agent. | [witness.ai](https://witness.ai/protect/), 12 Nov 2026 | | [Lasso Security](/platforms/lasso-security/) | partial | Credentials in MCP tool calls are detected and masked. | [lasso.security](https://www.lasso.security/use-cases/mcp-security), 12 Nov 2026 | | [Aim Security](/platforms/aim-security/) | none found | None found on 12 Nov 2026. | none found on 12 Nov 2026 | | [Koi](/platforms/koi/) | none found | No credential-access blocking documented. | [paloaltonetworks.com](https://www.paloaltonetworks.com/cortex/agentic-endpoint-security), 12 Nov 2026 | | [LayerX](/platforms/layerx/) | none found | None found on 12 Nov 2026. | none found on 12 Nov 2026 | | [Palo Alto Networks](/platforms/palo-alto-networks/) | none found | No credential-access blocking documented. | [paloaltonetworks.com](https://www.paloaltonetworks.com/prisma/prisma-ai-runtime-security), 12 Nov 2026 | Ordered by documentation level, then by the site ranking. Read 13 Nov 2026. Related: [Action Control Matrix](/action-control-matrix/), [how a cell is read](/action-control-matrix/methodology/). **About Bay.** Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). ## The other agent actions - [File read](/actions/file-read/) - [Shell command](/actions/shell-command/) - [Network call](/actions/network-call/) - [Data paste](/actions/data-paste/) - [Package install](/actions/package-install/) ## Sources - www.bay.io/platform/ai-endpoint-security: https://www.bay.io/platform/ai-endpoint-security (read 12 Nov 2026) - softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control: https://softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control (read 12 Nov 2026) - zenity.io/blog/coding-agent-attack-surface: https://zenity.io/blog/coding-agent-attack-surface (read 12 Nov 2026) - zenity.io/use-cases/agent-type/coding-personal-agents: https://zenity.io/use-cases/agent-type/coding-personal-agents (read 12 Nov 2026) - www.knostic.ai/faq: https://www.knostic.ai/faq (read 12 Nov 2026) - www.knostic.ai/kirin: https://www.knostic.ai/kirin (read 12 Nov 2026) - www.pillar.security/platform/agentic-endpoint: https://www.pillar.security/platform/agentic-endpoint (read 12 Nov 2026) - pulse2.com/vanta-to-acquire-pillar-security-to-expand-ai-agent-protection-as-customer-base-surpasses-20000: https://pulse2.com/vanta-to-acquire-pillar-security-to-expand-ai-agent-protection-as-customer-base-surpasses-20000/ (read 12 Nov 2026) - learn.microsoft.com/en-us/defender-endpoint/ai-agent-runtime-protection-overview: https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-runtime-protection-overview (read 12 Nov 2026) - www.harmonic.security/pricing: https://www.harmonic.security/pricing (read 12 Nov 2026) - www.crowdstrike.com/en-us/press-releases/crowdstrike-announces-general-availability-of-falcon-ai-detection-and-response: https://www.crowdstrike.com/en-us/press-releases/crowdstrike-announces-general-availability-of-falcon-ai-detection-and-response/ (read 12 Nov 2026) - www.crowdstrike.com/en-us/blog/crowdstrike-advances-ai-security-and-safety-falcon-guardian: https://www.crowdstrike.com/en-us/blog/crowdstrike-advances-ai-security-and-safety-falcon-guardian/ (read 12 Nov 2026) - noma.security/blog/noma-brings-agent-security-to-every-employee-endpoint: https://noma.security/blog/noma-brings-agent-security-to-every-employee-endpoint (read 12 Nov 2026) - noma.security/blog/securing-the-agentic-frontier-noma-unveils-the-first-real-time-agent-runtime-security-for-cursor: https://noma.security/blog/securing-the-agentic-frontier-noma-unveils-the-first-real-time-agent-runtime-security-for-cursor (read 12 Nov 2026) - github.com/Noma-Security/noma-marketplace/blob/main/guardrails/README.md: https://github.com/Noma-Security/noma-marketplace/blob/main/guardrails/README.md (read 12 Nov 2026) - www.island.io/product/ai: https://www.island.io/product/ai (read 12 Nov 2026) - www.island.io/press/island-brings-the-enterprise-browser-to-ai-agents: https://www.island.io/press/island-brings-the-enterprise-browser-to-ai-agents (read 12 Nov 2026) - www.sentinelone.com/press/sentinelone-launches-singularity-agentic-detection-and-response-to-stop-rogue-and-malicious-ai-agents-at-machine-speed: https://www.sentinelone.com/press/sentinelone-launches-singularity-agentic-detection-and-response-to-stop-rogue-and-malicious-ai-agents-at-machine-speed/ (read 12 Nov 2026) - witness.ai/protect: https://witness.ai/protect/ (read 12 Nov 2026) - www.lasso.security/use-cases/mcp-security: https://www.lasso.security/use-cases/mcp-security (read 12 Nov 2026) - www.paloaltonetworks.com/cortex/agentic-endpoint-security (checked, nothing found): https://www.paloaltonetworks.com/cortex/agentic-endpoint-security (read 12 Nov 2026) - www.paloaltonetworks.com/prisma/prisma-ai-runtime-security (checked, nothing found): https://www.paloaltonetworks.com/prisma/prisma-ai-runtime-security (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/actions/credential-access/ --- # 10 of 16 tools document stopping data paste before it completes AGENT ACTION, read 13 Nov 2026 **The short answer.** 10 of 16 AI endpoint security tools document stopping data paste, 5 document it in part, and nothing was found for 1 on 13 Nov 2026. Notes and sources for all 16 tools sit in the table below. Rows are ordered by documentation level, then by the site ranking. ## Which tools can stop a data paste? 10 of 16 tools document stopping data paste before it completes: Microsoft, Harmonic Security, Aim Security, CrowdStrike, Island, SentinelOne, WitnessAI, Lasso Security, LayerX and Palo Alto Networks. 5 more do in part (Bay, Noma Security, Zenity, Knostic and Koi). ## What this action is A user or agent pasting or uploading sensitive text into an AI tool. ## Tool by tool | Tool | Level | Note | Source | | --- | --- | --- | --- | | [Microsoft](/platforms/microsoft/) | documented | Endpoint DLP detects paste to restricted domains and blocks uploads from unallowed browsers. | [learn.microsoft.com](https://learn.microsoft.com/en-us/purview/endpoint-dlp-learn-about), 12 Nov 2026 | | [Harmonic Security](/platforms/harmonic-security/) | documented | Prompts with sensitive data can be blocked in browser, coding agents and desktop apps. | [docs.harmonicsecurity.app](https://docs.harmonicsecurity.app/portal-guides/detection-and-response/sensitive-data-controls-for-coding-agents-and-desktop-applications), 12 Nov 2026 | | [Aim Security](/platforms/aim-security/) | documented | Blocks sensitive prompts at send for desktop AI and can monitor, anonymize or block in the browser. | [catonetworks.com](https://www.catonetworks.com/platform/cato-ai-endpoint/), 12 Nov 2026 | | [CrowdStrike](/platforms/crowdstrike/) | documented | Sensitive data blocked before reaching models; Seraphic enforces paste and upload policy in the browser. | [crowdstrike.com](https://www.crowdstrike.com/en-us/press-releases/crowdstrike-announces-general-availability-of-falcon-ai-detection-and-response/), 12 Nov 2026 | | [Island](/platforms/island/) | documented | Clipboard, file and USB DLP on the endpoint; prompt and upload DLP in the browser. | [island.io](https://www.island.io/enterprise-desktop), 12 Nov 2026 | | [SentinelOne](/platforms/sentinelone/) | documented | Policy rules redact, tokenize or block sensitive prompts. | [sentinelone.com](https://www.sentinelone.com/blog/a-new-chapter-for-ai-and-cybersecurity-sentinelone-acquires-prompt-security/), 12 Nov 2026 | | [WitnessAI](/platforms/witnessai/) | documented | Redacts sensitive data in prompts in real time. | [witness.ai](https://witness.ai/), 12 Nov 2026 | | [Lasso Security](/platforms/lasso-security/) | documented | Prompts and uploads with PII or PCI are masked before reaching third-party models. | [lasso.security](https://www.lasso.security/platform/ai-usage-control), 12 Nov 2026 | | [LayerX](/platforms/layerx/) | documented | DLP over typed input, copy-paste and uploads with block or redact before data leaves. | [akamai.com](https://www.akamai.com/products/akamai-workforce-protector), 12 Nov 2026 | | [Palo Alto Networks](/platforms/palo-alto-networks/) | documented | Prisma Browser restricts copy-paste and uploads to unsanctioned AI apps and redacts prompts. | [paloaltonetworks.com](https://www.paloaltonetworks.com/sase/prisma-browser), 12 Nov 2026 | | [Bay](/platforms/bay/) | partial | Bay says browser guardrails keep personal data, secrets and source code out of prompts; detail on paste or upload blocking not found. | [bay.io](https://www.bay.io/platform/browser?utm_source=aiendpointsecurityranked.com&utm_medium=referral&utm_campaign=action), 12 Nov 2026 | | [Noma Security](/platforms/noma-security/) | partial | Inline masking of sensitive data is documented for agent traffic; no paste-into-AI-tool control found. | [noma.security](https://noma.security/blog/noma-brings-agent-security-to-every-employee-endpoint), 12 Nov 2026 | | [Zenity](/platforms/zenity/) | partial | Governance controls on data destinations; no paste or upload-to-chat control documented for desktop AI apps. | [zenity.io](https://zenity.io/use-cases/agent-type/coding-personal-agents), 12 Nov 2026 | | [Knostic](/platforms/knostic/) | partial | Secrets in prompts inside agents blocked; browser chatbot use is visibility-only. | [knostic.ai](https://www.knostic.ai/kirin), 12 Nov 2026 | | [Koi](/platforms/koi/) | partial | Blocks data exfiltration across AI interactions via Prisma AIRS. | [paloaltonetworks.com](https://www.paloaltonetworks.com/cortex/agentic-endpoint-security.md), 12 Nov 2026 | | [Pillar Security](/platforms/pillar-security/) | none found | Nothing found on 12 Nov 2026 on paste or upload controls for AI tools. | none found on 12 Nov 2026 | Ordered by documentation level, then by the site ranking. Read 13 Nov 2026. Related: [Action Control Matrix](/action-control-matrix/), [how a cell is read](/action-control-matrix/methodology/). **About Bay.** Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). ## The other agent actions - [File read](/actions/file-read/) - [Shell command](/actions/shell-command/) - [Network call](/actions/network-call/) - [Credential access](/actions/credential-access/) - [Package install](/actions/package-install/) ## Sources - learn.microsoft.com/en-us/purview/endpoint-dlp-learn-about: https://learn.microsoft.com/en-us/purview/endpoint-dlp-learn-about (read 12 Nov 2026) - docs.harmonicsecurity.app/portal-guides/detection-and-response/sensitive-data-controls-for-coding-agents-and-desktop-applications: https://docs.harmonicsecurity.app/portal-guides/detection-and-response/sensitive-data-controls-for-coding-agents-and-desktop-applications (read 12 Nov 2026) - www.harmonic.security/pricing: https://www.harmonic.security/pricing (read 12 Nov 2026) - www.catonetworks.com/platform/cato-ai-endpoint: https://www.catonetworks.com/platform/cato-ai-endpoint/ (read 12 Nov 2026) - www.crowdstrike.com/en-us/press-releases/crowdstrike-announces-general-availability-of-falcon-ai-detection-and-response: https://www.crowdstrike.com/en-us/press-releases/crowdstrike-announces-general-availability-of-falcon-ai-detection-and-response/ (read 12 Nov 2026) - www.crowdstrike.com/en-us/blog/crowdstrike-brings-enterprise-browser-security-to-falcon-platform: https://www.crowdstrike.com/en-us/blog/crowdstrike-brings-enterprise-browser-security-to-falcon-platform/ (read 12 Nov 2026) - www.island.io/enterprise-desktop: https://www.island.io/enterprise-desktop (read 12 Nov 2026) - www.island.io/blog/ai-runtime-protection-governing-what-agents-do-not-just-what-theyre-told: https://www.island.io/blog/ai-runtime-protection-governing-what-agents-do-not-just-what-theyre-told (read 12 Nov 2026) - www.securityweek.com/island-raises-400-million-at-6-4-billion-valuation: https://www.securityweek.com/island-raises-400-million-at-6-4-billion-valuation/ (read 12 Nov 2026) - betanews.com/2025/12/17/governing-ai-where-work-actually-happens-qa: https://betanews.com/2025/12/17/governing-ai-where-work-actually-happens-qa/ (read 12 Nov 2026) - www.sentinelone.com/blog/a-new-chapter-for-ai-and-cybersecurity-sentinelone-acquires-prompt-security: https://www.sentinelone.com/blog/a-new-chapter-for-ai-and-cybersecurity-sentinelone-acquires-prompt-security/ (read 12 Nov 2026) - www.sentinelone.com/platform/prompt-security-ai-usage-control: https://www.sentinelone.com/platform/prompt-security-ai-usage-control/ (read 13 Nov 2026) - witness.ai: https://witness.ai/ (read 12 Nov 2026) - witness.ai/protect: https://witness.ai/protect/ (read 12 Nov 2026) - www.lasso.security/platform/ai-usage-control: https://www.lasso.security/platform/ai-usage-control (read 12 Nov 2026) - www.akamai.com/products/akamai-workforce-protector: https://www.akamai.com/products/akamai-workforce-protector (read 12 Nov 2026) - www.akamai.com/solutions/ai-usage-control: https://www.akamai.com/solutions/ai-usage-control (read 12 Nov 2026) - www.paloaltonetworks.com/sase/prisma-browser: https://www.paloaltonetworks.com/sase/prisma-browser (read 12 Nov 2026) - www.bay.io/platform/browser: https://www.bay.io/platform/browser (read 12 Nov 2026) - noma.security/blog/noma-brings-agent-security-to-every-employee-endpoint: https://noma.security/blog/noma-brings-agent-security-to-every-employee-endpoint (read 12 Nov 2026) - zenity.io/use-cases/agent-type/coding-personal-agents: https://zenity.io/use-cases/agent-type/coding-personal-agents (read 12 Nov 2026) - zenity.io/platform/ai-detection-and-response: https://zenity.io/platform/ai-detection-and-response (read 12 Nov 2026) - www.knostic.ai/kirin: https://www.knostic.ai/kirin (read 12 Nov 2026) - www.paloaltonetworks.com/cortex/agentic-endpoint-security.md: https://www.paloaltonetworks.com/cortex/agentic-endpoint-security.md (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/actions/data-paste/ --- # 1 of 16 tools document stopping package install before it completes AGENT ACTION, read 13 Nov 2026 **The short answer.** 1 of 16 AI endpoint security tools document stopping package install, 6 document it in part, and nothing was found for 9 on 13 Nov 2026. Notes and sources for all 16 tools sit in the table below. Rows are ordered by documentation level, then by the site ranking. ## Which tools can block a package install? 1 of 16 tools document stopping package install before it completes: CrowdStrike. 6 more do in part (Bay, Island, Zenity, Knostic, Koi and LayerX). ## What this action is An agent installing a package, plugin, skill or extension. ## Tool by tool | Tool | Level | Note | Source | | --- | --- | --- | --- | | [CrowdStrike](/platforms/crowdstrike/) | documented | Sensor blocks malicious npm and PyPI packages before embedded code can execute. | [crowdstrike.com](https://www.crowdstrike.com/en-us/blog/crowdstrike-extends-endpoint-security-to-stop-supply-chain-attacks/), 12 Nov 2026 | | [Bay](/platforms/bay/) | partial | Analyst profile reports approve, block and uninstall controls for packages and extensions; vendor-reported. | [softwareanalyst.substack.com](https://softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control), 12 Nov 2026 | | [Island](/platforms/island/) | partial | Code packages are scanned and flagged installs can be removed; pre-install blocking is not specified. | [island.io](https://www.island.io/product/ai), 12 Nov 2026 | | [Zenity](/platforms/zenity/) | partial | Malicious package install prevention named on the product page; details sit in JS-only docs. | [zenity.io](https://zenity.io/use-cases/agent-type/coding-personal-agents), 12 Nov 2026 | | [Knostic](/platforms/knostic/) | partial | Dependency inspection and malicious package alerts; blocking in Enforce is claimed generally. | [knostic.ai](https://www.knostic.ai/kirin), 12 Nov 2026 | | [Koi](/platforms/koi/) | partial | Stops risky downloads and offers one-click removal of insecure packages and extensions. | [paloaltonetworks.com](https://www.paloaltonetworks.com/resources/datasheets/cortex-agentic-endpoint-security-at-a-glance), 12 Nov 2026 | | [LayerX](/platforms/layerx/) | partial | Blocks malicious or overprivileged browser extensions; code package installs are not covered. | [akamai.com](https://www.akamai.com/products/akamai-workforce-protector), 12 Nov 2026 | | [Microsoft](/platforms/microsoft/) | none found | None found on 12 Nov 2026 in Defender AI agent docs. | none found on 12 Nov 2026 | | [Harmonic Security](/platforms/harmonic-security/) | none found | No package install controls found. | [docs.harmonicsecurity.app](https://docs.harmonicsecurity.app/portal-guides/detection-and-response/sensitive-data-controls-for-coding-agents-and-desktop-applications), 12 Nov 2026 | | [Aim Security](/platforms/aim-security/) | none found | None found on 12 Nov 2026. | none found on 12 Nov 2026 | | [Noma Security](/platforms/noma-security/) | none found | Nothing found on 12 Nov 2026 on blocking package installs by agents. | none found on 12 Nov 2026 | | [SentinelOne](/platforms/sentinelone/) | none found | None found on 12 Nov 2026. | none found on 12 Nov 2026 | | [WitnessAI](/platforms/witnessai/) | none found | No package-install blocking documented. | none found on 12 Nov 2026 | | [Lasso Security](/platforms/lasso-security/) | none found | No package install controls found. | [lasso.security](https://www.lasso.security/use-cases/ai-coding-assistants), 12 Nov 2026 | | [Pillar Security](/platforms/pillar-security/) | none found | Nothing found on 12 Nov 2026 on package-install blocking. | none found on 12 Nov 2026 | | [Palo Alto Networks](/platforms/palo-alto-networks/) | none found | None found on 12 Nov 2026 for Prisma AIRS or Prisma Browser; Cortex Agentic Endpoint Security (Koi) coverage is scored on the Koi profile. | none found on 12 Nov 2026 | Ordered by documentation level, then by the site ranking. Read 13 Nov 2026. Related: [Action Control Matrix](/action-control-matrix/), [how a cell is read](/action-control-matrix/methodology/). **About Bay.** Bay ranks 2nd of 16 at 71.5, behind Microsoft at 75.0. Microsoft's lead over Bay comes from transparency (7 to 1) and policy and governance (9 to 6). ## The other agent actions - [File read](/actions/file-read/) - [Shell command](/actions/shell-command/) - [Network call](/actions/network-call/) - [Credential access](/actions/credential-access/) - [Data paste](/actions/data-paste/) ## Sources - www.crowdstrike.com/en-us/blog/crowdstrike-extends-endpoint-security-to-stop-supply-chain-attacks: https://www.crowdstrike.com/en-us/blog/crowdstrike-extends-endpoint-security-to-stop-supply-chain-attacks/ (read 12 Nov 2026) - softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control: https://softwareanalyst.substack.com/p/the-ciso-guide-to-endpoint-control (read 12 Nov 2026) - www.island.io/product/ai: https://www.island.io/product/ai (read 12 Nov 2026) - www.island.io/blog/expanding-island-to-govern-your-ai-workforce: https://www.island.io/blog/expanding-island-to-govern-your-ai-workforce (read 12 Nov 2026) - zenity.io/use-cases/agent-type/coding-personal-agents: https://zenity.io/use-cases/agent-type/coding-personal-agents (read 12 Nov 2026) - www.knostic.ai/kirin: https://www.knostic.ai/kirin (read 12 Nov 2026) - www.paloaltonetworks.com/resources/datasheets/cortex-agentic-endpoint-security-at-a-glance: https://www.paloaltonetworks.com/resources/datasheets/cortex-agentic-endpoint-security-at-a-glance (read 12 Nov 2026) - www.paloaltonetworks.com/cortex/agentic-endpoint-security.md: https://www.paloaltonetworks.com/cortex/agentic-endpoint-security.md (read 12 Nov 2026) - www.calcalistech.com/ctechnews/article/xpz78znuc: https://www.calcalistech.com/ctechnews/article/xpz78znuc (read 12 Nov 2026) - www.securityweek.com/palo-alto-networks-to-acquire-koi-in-reported-400-million-transaction: https://www.securityweek.com/palo-alto-networks-to-acquire-koi-in-reported-400-million-transaction/ (read 12 Nov 2026) - www.akamai.com/products/akamai-workforce-protector: https://www.akamai.com/products/akamai-workforce-protector (read 12 Nov 2026) - docs.harmonicsecurity.app/portal-guides/detection-and-response/sensitive-data-controls-for-coding-agents-and-desktop-applications (checked, nothing found): https://docs.harmonicsecurity.app/portal-guides/detection-and-response/sensitive-data-controls-for-coding-agents-and-desktop-applications (read 12 Nov 2026) - www.lasso.security/use-cases/ai-coding-assistants (checked, nothing found): https://www.lasso.security/use-cases/ai-coding-assistants (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/actions/package-install/ --- # Six actions an agent can take, and how many tools can stop each AGENT ACTIONS, read 13 Nov 2026 **The short answer.** The Action Control Matrix splits into 6 actions: file read (4 of 16), shell command (7 of 16), network call (5 of 16), credential access (4 of 16), data paste (10 of 16), package install (1 of 16). The counts are tools with a documented capability on 13 Nov 2026. - [File read](/actions/file-read/): 4 of 16 tools documented. An agent opening a local file, including files outside the project it was started in. - [Shell command](/actions/shell-command/): 7 of 16 tools documented. An agent running a command in a shell on the endpoint. - [Network call](/actions/network-call/): 5 of 16 tools documented. An agent sending a request to an outside host or service. - [Credential access](/actions/credential-access/): 4 of 16 tools documented. An agent reading keys, tokens, cookies or password stores. - [Data paste](/actions/data-paste/): 10 of 16 tools documented. A user or agent pasting or uploading sensitive text into an AI tool. - [Package install](/actions/package-install/): 1 of 16 tools documented. An agent installing a package, plugin, skill or extension. ## Sources - Action Control Matrix: https://aiendpointsecurityranked.com/action-control-matrix/ (read 13 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/actions/ --- # Claude Code is safe to run on a laptop only after you turn on the sandbox and lock the managed settings GUIDE, read 12 Nov 2026 **The short answer.** Claude Code ships a permission system, an optional OS sandbox (Seatbelt on macOS, bubblewrap on Linux), hooks, MCP allowlists and OpenTelemetry export. Anthropic's docs say the sandbox is off by default and the working directory limit is a prompt, not a wall. Here is what to set, in order, and what remains. ## How do I secure Claude Code on developer laptops? Anthropic's security page describes Claude Code as permission based. In Manual mode the tool starts with read-only permissions and asks before it edits files, runs tests or executes commands. A built-in set of read-only commands such as ls, cat and git status runs without asking. Anthropic also documents an Auto mode, in which a separate classifier model reviews actions instead of the user, and says Auto mode is the built-in starting mode for interactive terminal and VS Code sessions. ## The default prompts reduce risk but do not form a boundary The same page is direct about the limit. The working directory boundary is a permission prompt, so a Bash command a developer approves can still write anywhere the user account can. Anthropic's own wording on untrusted content is that no system is completely immune to attack, and it recommends virtual machines for scripts and tool calls that touch external services. That framing sets the job. The prompts catch the obvious cases and train developers to read commands. Three controls turn that into something an administrator can rely on: the sandbox for operating system level limits, managed settings so developers cannot loosen them, and telemetry so you can see what ran. ## Permission rules match command text, so pair them with the sandbox Claude Code evaluates deny rules first. Anthropic's permissions page says a broad deny rule such as Bash(aws *) blocks every matching call, even when a narrower allow rule also matches, and that an allow rule cannot carve an exception out of a deny rule. Matching ask rules also win over allow rules. A bare tool name in a deny rule, for example Bash, removes the tool from Claude's context entirely. Pattern rules behave differently, and the docs list what they miss. A deny rule on curl does not match the same program invoked by path or inside sh -c, so Anthropic tells teams to pair it with the sandbox network allowlist when the restriction must hold. File rules have a similar edge. Read and Edit deny rules apply to the built-in file tools, to file commands Claude Code recognizes in Bash such as cat, head, tail, sed and tee, and to redirection targets. They do not apply to a command that reads files without naming them, such as a recursive grep run from the directory that holds the file. - **Protect secrets files.** Add Read deny rules for paths such as Read(./.env) and Read(./secrets/**). A .claudeignore file has no effect, per the permissions page. - **Disable the unsafe modes.** Set permissions.disableBypassPermissionsMode and permissions.disableAutoMode to "disable" where you do not want them. Anthropic says these work best in managed settings, where they cannot be overridden. - **Stop broad allow lists from accumulating.** Claude Code merges permissions.allow across scopes unless the organization sets allowManagedPermissionRulesOnly. ## The sandbox is off by default and covers Bash, not the built-in tools Anthropic's sandboxing page states that the sandbox is off by default. A developer turns it on with /sandbox, or an administrator sets sandbox.enabled to true. On macOS it uses the Seatbelt framework. On Linux and WSL2 it relies on bubblewrap and socat, and the /sandbox panel shows what is missing. Inside the sandbox, the docs describe a filesystem that reads most of the machine by default, including credential files such as ~/.ssh and ~/.aws/credentials, unless filesystem.denyRead says otherwise. Writes go to the working directory, a per-user temp directory and any added directories. Network traffic has no direct route out: connections pass through a proxy on the machine that checks each host against network.allowedDomains, which starts empty. Two caveats matter for a security review. First, the sandbox wraps Bash commands and their child processes. Anthropic says the built-in Read, Edit, Write, WebFetch and WebSearch tools follow permission rules instead, so a denyRead entry does not stop the Read tool and allowedDomains does not limit WebFetch. Second, if the platform cannot support the sandbox or the dependency check fails, Claude Code runs commands without sandboxing by default. - **Fail closed.** Set sandbox.failIfUnavailable to true so a missing dependency stops startup instead of silently dropping isolation. Anthropic describes this as the setting for managed deployments that treat sandboxing as a security gate. - **Remove the unsandboxed retry.** Set allowUnsandboxedCommands to false. With the retry disabled, Claude Code ignores the dangerouslyDisableSandbox parameter. - **Deny credential reads.** Add ~/.ssh, ~/.aws and similar paths to filesystem.denyRead, since the default view of the machine is wide. - **Start the domain list small.** Leave network.allowedDomains empty and add hosts as builds require them. ## Managed settings are the control developers cannot edit Anthropic's settings page lists managed settings as the highest scope. They arrive as a managed-settings.json file, an MDM policy or server-managed settings from the claude.ai console. The page states that nothing in a user's own settings files or in the --settings flag overrides a managed key. Managed settings that arrive through MDM or the console reach a running session on a schedule rather than on save. A practical baseline for a laptop fleet uses these keys, all named in Anthropic's documentation: sandbox.enabled with failIfUnavailable and allowUnsandboxedCommands set as above, permissions.disableBypassPermissionsMode, permissions.disableAutoMode where classifier approval is not accepted, allowManagedPermissionRulesOnly, allowManagedHooksOnly, and the MCP allowlist keys covered below. Anthropic notes that a managed model value is a starting default, not a lock; the locks are availableModels and deniedModels. One scope gap is worth stating. A managed-settings.json file or MDM profile on a laptop does not reach a cloud session. Only server-managed settings do. If developers also use cloud sessions, plan for both delivery paths. ## Hooks give a pre-execution decision point, and managed policy can restrict who writes them Claude Code hooks run at defined points in the agent loop. Anthropic's hooks reference lists PreToolUse as firing before a tool call executes and able to block it. Its worked example reads the proposed command on standard input and returns a permissionDecision of deny with a reason. That is the vendor-supported place to add an organization check that sees the command before it runs. Because hooks are code that runs on the developer's machine, Anthropic lets administrators set allowManagedHooksOnly in managed settings to restrict which hooks run. Its security page also points to ConfigChange hooks to audit or block settings changes during a session. The settings page adds a limit: ConfigChange runs for settings-file changes it detects, not for managed settings that arrive from MDM or the console. ## MCP servers and repository files can add capability that the repository review misses Project-scoped MCP servers live in .mcp.json, which teams can check into source control. Anthropic's security page warns that reviewing .mcp.json does not show every server a session can load, because servers at other scopes, claude.ai connectors and plugins can add more. It also says Anthropic reviews connectors against listing criteria but does not security-audit or manage any MCP server. Claude Code shows a workspace trust dialog when it starts in a folder the user has not trusted, and servers in .mcp.json have their own approval prompt. The page notes that a -p (non-interactive) session shows neither prompt. When Claude Code starts directly in a home directory, trust is held for the session only. The MCP documentation describes the organization controls: a fixed server set through managed-mcp.json, servers provided to every user through managedMcpServers, and restriction through allowedMcpServers and deniedMcpServers. For a developer who must stay inside a known set, the --strict-mcp-config flag makes a session use only servers passed with --mcp-config. ## OpenTelemetry export shows what ran, if you set it in managed settings Anthropic's monitoring page describes OpenTelemetry export that is off until CLAUDE_CODE_ENABLE_TELEMETRY is set to 1. Events include claude_code.user_prompt, claude_code.tool_decision and claude_code.tool_result, linked by a tool_use_id so a decision can be joined to its result. A prompt.id value groups everything triggered by one prompt. Claude Code ignores the exporter variables in a repository's .claude/settings.json, so a cloned repository cannot turn telemetry on, change its destination or capture content. Set them in managed settings. Detail is opt-in: OTEL_LOG_TOOL_DETAILS adds Bash commands and MCP tool names, and OTEL_LOG_RAW_API_BODIES, disabled by default, emits full request and response bodies including the whole conversation. Decide deliberately whether command text and prompts belong in your log store. ## An endpoint control covers what the tool's own settings cannot see Claude Code controls its own behavior. An endpoint control sees the machine around it: other agents, other shells, the packages that arrive, and processes Claude Code did not start. Anthropic's devcontainer page shows why the boundary matters. With --dangerously-skip-permissions, it says, a dev container does not prevent a malicious project from exfiltrating anything reachable inside the container, including the Claude Code credentials in ~/.claude, and it pairs the flag with a firewall script for egress. Several vendors say they attach to Claude Code's own hook interface to decide before an action runs. CrowdStrike says agent-level hooks for Claude Code, Codex, Copilot and Cursor can inspect prompts and block malicious activity before harmful actions run. Microsoft says agent-native inspection uses vendor event interfaces for Claude Code, Codex CLI and GitHub Copilot CLI to apply audit or block decisions. Pillar says its runtime component sits in the agent's own pre-tool hook for Claude Code, Codex CLI and Cursor. Microsoft also lists a preview known issue: Claude Code may keep processing a tool response in Block mode, and post-response inspection cannot undo completed actions. These are the vendors' claims, read on 12 Nov 2026. Weigh any of them against what the managed settings already deliver. A hook that duplicates a deny rule adds little. A hook that applies a central decision to every agent on the machine, with a log outside the developer's reach, adds visibility Claude Code's own telemetry cannot provide. ## Where this sits in the index - [Agent Surface Map](/agent-surface-map/): Which of six agent surfaces each of 16 tools sees - [Action Control Matrix](/action-control-matrix/): Which of six agent actions each tool can stop - [The ranked index](/ranking/): 16 tools on seven published criteria ## Related guides - [Securing Cursor on developer machines](/guides/cursor-security/) - [Securing GitHub Copilot agent mode and the cloud agent](/guides/github-copilot-agent-mode-security/) - [Agent surfaces explained](/guides/agent-surfaces-explained/) ## Sources - Anthropic: Claude Code docs, security: https://code.claude.com/docs/en/security (read 12 Nov 2026) - Anthropic: Claude Code docs, permissions: https://code.claude.com/docs/en/permissions (read 12 Nov 2026) - Anthropic: Claude Code docs, sandboxing: https://code.claude.com/docs/en/sandboxing (read 12 Nov 2026) - Anthropic: Claude Code docs, settings: https://code.claude.com/docs/en/settings (read 12 Nov 2026) - Anthropic: Claude Code docs, hooks: https://code.claude.com/docs/en/hooks (read 12 Nov 2026) - Anthropic: Claude Code docs, development containers: https://code.claude.com/docs/en/devcontainer (read 12 Nov 2026) - Anthropic: Claude Code docs, MCP: https://code.claude.com/docs/en/mcp (read 12 Nov 2026) - Anthropic: Claude Code docs, monitoring usage: https://code.claude.com/docs/en/monitoring-usage (read 12 Nov 2026) - CrowdStrike: Falcon Guardian blog post: https://www.crowdstrike.com/en-us/blog/crowdstrike-advances-ai-security-and-safety-falcon-guardian/ (read 12 Nov 2026) - Microsoft: Defender AI agent runtime protection overview: https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-runtime-protection-overview (read 12 Nov 2026) - Pillar Security: Agentic Endpoint platform page: https://www.pillar.security/platform/agentic-endpoint (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/guides/claude-code-security/ --- # Cursor's guardrails are best-effort, so pick Auto-review or Allowlist, never Run Everything GUIDE, read 12 Nov 2026 **The short answer.** Cursor's agent reads files and edits workspace files without approval, asks before shell commands and MCP calls, and offers three Run Modes plus a sandbox. Cursor calls its guardrails best-effort, not a hard boundary. Privacy Mode is default for Enterprise teams. Settings, hooks and audit log limits are covered below. ## Cursor's defaults let the agent read and edit freely and ask before it acts outside the editor Cursor's agent security page lays out the default behavior. Reading files and searching code do not require approval. Agents can modify workspace files without approval, except configuration files such as workspace settings, which need approval first. Changes save to disk immediately, and Cursor warns that with auto-reload enabled, agent changes might run before a developer can review them. Terminal commands need approval by default. All MCP connections need approval, and after a connection is approved each tool call still needs individual approval unless the tool is on an MCP allowlist. The page says the agent cannot make arbitrary network requests with default settings; Cursor lists the destinations its tools contact. Cursor is candid about the strength of these controls. It describes Run Modes as best-effort guardrails rather than a hard security boundary, and recommends keeping the defaults enabled. For a security reader that sentence decides the design: treat the prompts as a speed bump, and put hard limits in the sandbox, the managed policy and an external control. ## Run Modes decide how much autonomy the agent gets Cursor's Run Modes page, found under Settings, Agents, Approvals and Execution, lists three modes. Auto-review runs allowlisted calls immediately, sandboxes other shell commands when possible, and sends calls that cannot use the sandbox to a classifier. Allowlist runs only the actions on a list and is described as deterministic. Run Everything runs every tool call with no sandbox and no classifier. The classifier deserves a closer read. Cursor says it runs on the Cursor backend and can make read-only ReadFile, Grep, Glob and ListDir calls on the developer's machine, for example to read a script a command runs. It runs on small Cursor-managed models, which the page names as Gemini 3.5 Flash Lite with Haiku as the fallback. Cursor states plainly that the classifier can make mistakes in either direction. If a sandboxed command fails on a sandbox restriction, the agent can rerun it outside the sandbox, and the classifier reviews that rerun. Teams can steer the classifier in plain English through permissions.json, either at ~/.cursor/permissions.json or in a project's .cursor/permissions.json, with allow_instructions and block_instructions. A team configuration defined in the dashboard takes priority and Cursor then ignores the user and project files. - **Choose per risk.** Auto-review fits most developers. Allowlist fits repositories where you want a short fixed set of trusted actions. - **Keep a model available.** Cursor says blocking Claude 4.5 Haiku in enterprise model controls can disable Auto-review, and members then fall back to Allowlist. - **Write block instructions for the commands that matter locally.** The page's own examples are cloud CLI commands and anything that modifies Kubernetes resources. ## The sandbox limits shell commands but lets some commands step outside it Per Cursor's Run Modes page, the sandbox runs terminal commands in a restricted environment. Commands get read and write access inside the workspace. Protected paths include .git/config, .git/hooks, .vscode, .cursorignore and sensitive Cursor configuration. Network is blocked by default, then opened by the network mode and by sandbox.json. Temporary files are writable unless the file disables them. Some commands need more access, such as full network access, writes outside the workspace or privileged operations. They cannot use the sandbox. Cursor says it indicates when a command runs outside the sandbox and asks for approval. sandbox.json can be set per user at ~/.cursor/sandbox.json or per project, with the project file taking priority, and Cursor states that team-admin policies and its hardcoded rules layer on top so local files cannot weaken them. On reads, Cursor documents a setting that many teams will want to change. Read Access defaults to System, which lets the agent read files outside the workspace without approval. Setting it to Workspace makes reads outside the workspace require approval unless the path is on a Read Allowlist. Read Access does not apply in Run Everything mode. For laptops that hold SSH keys and cloud credentials, Workspace is the safer choice, and it can also be set per machine or project with readBoundary in sandbox.json. ## MCP servers add tools that Cursor approves one call at a time Cursor's MCP page covers local and remote servers over stdio, SSE and streamable HTTP, with OAuth for remote servers. Team admins can distribute servers through a team marketplace, and these appear alongside personal and workspace servers. Static OAuth client credentials can be supplied in mcp.json for providers that do not support dynamic client registration, and values can come from environment variables rather than being written into the file. Cursor's security page links a separate set of MCP security considerations in its developer docs. The default approval of the connection and then each call is the main control. The weak point is the allowlist: once a developer pre-approves a tool, calls to it run without a prompt, so review which servers sit on team-distributed lists and what each tool can reach. Cursor workspace trust exists but is disabled by default, per the agent security page. When enabled, it prompts for normal or restricted mode on new workspaces, and restricted mode breaks AI features. Cursor's advice for untrusted repositories is to use a basic text editor, and it says organizations can enforce the trust setting through MDM. ## Privacy Mode governs where prompts and code go, and Cloud Agents are the exception Cursor's security page says Privacy Mode can be enabled in settings or by a team or enterprise admin, is available to free and Pro users, and is inherited by new team members. With it on, Cursor says it will not train on the data. Its data governance page says Privacy Mode is on by default for Enterprise teams, and that most models run under zero data retention agreements. That page also identifies the one feature that stores code. Cloud Agents are the only feature that requires Cursor to store code, and Cursor says that if a security policy prohibits code storage, the answer is not to enable Cloud Agents. A few models require data retention at the provider. For Enterprise customers and teams with Privacy Mode, Cursor requires admin approval before those models can be used. For corporate devices, Cursor documents an Allowed Team IDs policy that prevents users from logging into personal accounts, which might lack Privacy Mode. Cursor points to MDM deployment patterns for rolling out the policy. Cursor's security page lists ISO/IEC 27001:2022, ISO/IEC 42001:2023, AIUC-1 and SOC 2 Type II as held certifications. ## Hooks and audit logs cover different ground Cursor hooks are scripts that run before or after stages of the agent loop. The hooks page says they are spawned processes that communicate by JSON over stdio, are defined in hooks.json at the project or user level or installed through plugins, and can observe, block or modify behavior. Stages named on the page include beforeShellExecution, beforeMCPExecution, beforeReadFile, afterFileEdit and the generic preToolUse. The page also notes that Cursor can load hooks from third-party tools such as Claude Code. The audit log is narrower than many buyers assume. Cursor's compliance page says audit logs record who changed what in the team: sign-ins, membership, roles, API keys, settings and integrations. Payloads never contain prompts, agent output, generated code or credentials, and the logs are available on the Enterprise plan. For development activity, the same page tells customers to use hooks. A security team that wants a record of agent shell commands therefore has to build or buy a hook-based collector. ## An endpoint control adds the record and the decision Cursor leaves to hooks Cursor's own design leaves two gaps for a security team. The guardrails are best-effort by Cursor's account, and the audit log omits agent activity. Hooks are the sanctioned place to fill both, and several vendors build on them. Noma says it uses Cursor Agent Hooks for inline decisions before actions happen, aimed at destructive operations, credential leakage, indirect injection, tool misuse and workspace-trust code execution. Pillar says its runtime component sits in the pre-tool hook for Claude Code, Codex CLI and Cursor. Zenity says native hooks and its MCP gateway can block or modify a dangerous tool call before it executes in Cursor, among other coding agents. CrowdStrike lists Cursor among agents with hooks that can inspect prompts and block malicious activity. These are vendor claims, read on 12 Nov 2026. When you evaluate one, test three things Cursor's docs make relevant: whether it fires on beforeShellExecution and beforeMCPExecution, whether it logs the command text off the laptop, and whether it still has an effect when a developer chooses Run Everything, since that mode runs every call without checks. ## Where this sits in the index - [Agent Surface Map](/agent-surface-map/): Which of six agent surfaces each of 16 tools sees - [Action Control Matrix](/action-control-matrix/): Which of six agent actions each tool can stop - [The ranked index](/ranking/): 16 tools on seven published criteria ## Related guides - [Securing GitHub Copilot agent mode and the cloud agent](/guides/github-copilot-agent-mode-security/) - [Securing OpenAI Codex CLI on developer laptops](/guides/codex-cli-security/) - [Securing Claude Code on developer laptops](/guides/claude-code-security/) ## Sources - Cursor: security: https://cursor.com/security (read 12 Nov 2026) - Cursor: docs, agent security: https://cursor.com/docs/agent/security (read 12 Nov 2026) - Cursor: docs, run modes and sandboxing: https://cursor.com/docs/agent/security/run-modes (read 12 Nov 2026) - Cursor: docs, MCP: https://cursor.com/docs/mcp (read 12 Nov 2026) - Cursor: docs, hooks: https://cursor.com/docs/hooks (read 12 Nov 2026) - Cursor: docs, privacy and data governance: https://cursor.com/docs/enterprise/privacy-and-data-governance (read 12 Nov 2026) - Cursor: docs, compliance and monitoring: https://cursor.com/docs/enterprise/compliance-and-monitoring (read 12 Nov 2026) - Noma Security: runtime security for Cursor blog post: https://noma.security/blog/securing-the-agentic-frontier-noma-unveils-the-first-real-time-agent-runtime-security-for-cursor (read 12 Nov 2026) - Pillar Security: Agentic Endpoint platform page: https://www.pillar.security/platform/agentic-endpoint (read 12 Nov 2026) - Zenity: coding and personal agents use case: https://zenity.io/use-cases/agent-type/coding-personal-agents (read 12 Nov 2026) - CrowdStrike: Falcon Guardian blog post: https://www.crowdstrike.com/en-us/blog/crowdstrike-advances-ai-security-and-safety-falcon-guardian/ (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/guides/cursor-security/ --- # Copilot's local agent mode needs sandboxing turned on, and its cloud agent firewall is not a full control GUIDE, read 12 Nov 2026 **The short answer.** Copilot agent mode in VS Code approves terminal commands, MCP tools and URLs, but network access defaults to on (chat.agent.sandbox.network.allowNetwork is true). GitHub's cloud agent pushes only to a copilot/ branch, yet its firewall skips MCP server processes. Settings, managed policy and gaps are set out below. ## Copilot agent mode runs in two places with two different security models Developers use Copilot's agent in two ways, and the controls do not carry across. Agent mode in VS Code works on the laptop, in the developer's workspace, with the developer's credentials. GitHub's cloud agent works in a GitHub-hosted environment, opens a pull request and never touches the laptop. GitHub's current documentation calls the second one the Copilot cloud agent; this guide uses that name. The VS Code security documentation draws the line itself: managed compute is a different boundary from process-level agent sandboxing, and local sandbox or approval settings do not configure the cloud service. A security review therefore needs two checklists. The local one is about shell, files, MCP and credentials on the device. The cloud one is about branches, workflows and a network firewall. ## Local approvals come in four levels, and the top two remove the review Per the VS Code security page, a Copilot session has a permissions picker. Manual permissions uses the configured approval settings. Assisted permissions uses an LLM judge to evaluate each tool call and asks the developer when the judge does not approve it. Allow all auto-approves every tool call. Autopilot is an agent mode that auto-approves all tools and drives the agent to continue until the task is complete. The page recommends keeping auto-approval scoped to the session rather than the workspace or user level, which limits how long wider trust lasts. When a tool or terminal command is approved automatically, VS Code shows an information message with a link to the setting that enabled it, which helps a developer notice drift but does not alert an administrator. Untrusted projects are the other entry point. VS Code's guidance is to open them in restricted mode, which disables agents in that workspace until the project has been reviewed. Trust decisions cover the workspace, extension publishers, MCP servers and network domains, and MCP servers in .vscode/mcp.json and a workspace-root .mcp.json inherit Workspace Trust. ## Terminal commands need approval, and the sandbox does not restrict the network by default VS Code states that before executing terminal commands the agent requests explicit user approval. When terminal auto-approval is enabled, configurable per-command rules, including regex patterns, approve safe commands and prompt for potentially dangerous ones. Every subcommand in a compound command must match an approved rule. The docs add that terminal controls apply to Local harness sessions and configurations using the integrated terminal. Agent sandboxing is available on supported platforms and restricts file system access for shell commands the agent runs and their child processes. The security page is specific about its network default: outbound network access stays enabled because chat.agent.sandbox.network.allowNetwork defaults to true, and network access has to be configured separately. A team that turns on sandboxing and stops there has limited file access but not exfiltration paths. VS Code also documents a direct shell shortcut: a chat message that starts with ! runs the command immediately in the session's default shell without sending it to the agent or asking for approval. That is a developer action, not an agent action, but it is worth knowing when reading shell history. - **Turn on agent sandboxing** on supported platforms and set the network setting to match your egress policy. - **Keep terminal auto-approval rules short.** Review regex rules the way you review firewall rules, because an over-broad pattern approves everything it matches. - **Use restricted mode** for any repository the developer has not reviewed. ## File edits are reviewable only for edits made through the editor's own tools VS Code offers a setting, chat.tools.edits.autoApprove, with glob patterns such as "**/.env": false that require manual approval for edits to sensitive files. The security page recommends it for protecting those files. The review surface has a stated limit. Keep and Undo tracks only edits made through supported edit tools, so changes made through terminal commands need separate review. The docs also say that keeping an edit acknowledges an already saved change; it is not approval before execution. The review is after the fact, which is why the terminal approval and sandbox settings carry the real weight. ## MCP servers run with the developer's permissions and have an organization allowlist VS Code's MCP documentation says plainly that local MCP servers can run code on the machine and that tools can access services using the developer's permissions. It tells users to review the publisher, the command or endpoint and the requested access before installing or starting a server, and not to enter passwords or API keys in server input forms. The security page adds that sensitive input parameters are stored in VS Code's secure credentials store and that VS Code implements the MCP authorization specification for OAuth. For an organization, the enterprise page describes allowedMcpServers and deniedMcpServers in Copilot enterprise-managed settings. The allowlist limits which MCP servers developers can install and run, and the denylist blocks servers even when they match the allowlist. GitHub's example permits remote servers under one URL prefix while blocking one endpoint. These keys must be delivered through a managed-settings method, not through a developer's editor settings or mcp.json. ## Managed settings, device policy and GitHub account settings are three separate layers The VS Code enterprise page names three management solutions. Copilot enterprise-managed settings apply guardrails across supported clients such as VS Code and Copilot CLI, delivered as server-managed, MDM-managed or file-based settings, with team overrides for server-managed deployments. VS Code device policies are deployed through Group Policy, Microsoft Intune or macOS configuration profiles and override the corresponding user settings. GitHub organization and enterprise settings cover account-level controls such as Copilot access, model availability and content exclusions, and follow the signed-in GitHub account rather than the device. When enterprise-managed settings and a device policy configure the same control, the enterprise-managed setting wins and the values are not merged. Not every setting applies to every client, so check the supported-clients column in GitHub's reference. Developer: Policy Diagnostics in VS Code shows the effective value and its source, which is the quickest way to confirm a rollout reached a laptop. One interaction affects tool choice. The security page says Claude and Codex harnesses cannot enforce Copilot enterprise-managed settings, and they are disabled when enterprise-managed settings apply, including managed permission or sandbox settings. If your developers expect to use several agents inside VS Code, that is a decision to communicate before the policy lands. ## The cloud agent limits what it can push, but its firewall has stated gaps GitHub's risk documentation for the cloud agent lists its mitigations. It can push to a single branch: the pull request's branch when mentioned there, otherwise a new copilot/ branch. It cannot run git push directly, and branch protections and required checks still apply. Pull requests it creates are drafts that a human must review and merge. It cannot mark them ready for review, approve them or merge them, and the person who asked for the work cannot approve the pull request. By default, GitHub Actions workflows do not run until a user with write access clicks Approve and run workflows. GitHub also filters hidden characters, such as HTML comments, from issue and comment text before the agent sees it. Network access is limited by a firewall configurable at the organization and repository level, with a recommended allowlist enabled by default so the agent can download dependencies. GitHub is explicit about its limits. The firewall applies only to processes the agent starts through its Bash tool, not directly to MCP server processes or processes started in configured setup steps. It applies only inside the agent's environment. GitHub also says sophisticated attacks may bypass it and that it should not be considered a comprehensive security solution. When the firewall blocks a request, a warning appears in the pull request with the blocked address and the command. For automations that run on a schedule or on events, GitHub says input from untrusted users could become part of the prompt, so the owner chooses exactly which tools the automation may use. ## An endpoint control fills the local half that GitHub's controls leave to the device GitHub's cloud controls sit on GitHub's side, so an endpoint tool adds nothing there. The gap is local. Agent mode runs on the laptop, the sandbox network default is open, and the review surface does not track terminal changes. Vendors that say they cover Copilot at the endpoint include Microsoft, which says Defender collects installed VS Code extension data for agents like GitHub Copilot, and applies audit or block decisions through vendor event interfaces for GitHub Copilot CLI. CrowdStrike says hooks for GitHub Copilot can inspect prompts and block malicious activity before harmful actions run, and Zenity makes a similar pre-execution claim. Those are the vendors' statements, read on 12 Nov 2026. Ask each vendor which Copilot surface it covers. Microsoft's own Defender documentation separates VS Code extension collection from Copilot CLI inspection, and the two are not the same interface. ## Where this sits in the index - [Agent Surface Map](/agent-surface-map/): Which of six agent surfaces each of 16 tools sees - [Action Control Matrix](/action-control-matrix/): Which of six agent actions each tool can stop - [The ranked index](/ranking/): 16 tools on seven published criteria ## Related guides - [Securing OpenAI Codex CLI on developer laptops](/guides/codex-cli-security/) - [Securing Google Gemini CLI on developer laptops](/guides/gemini-cli-security/) - [Securing Cursor on developer machines](/guides/cursor-security/) ## Sources - Microsoft: VS Code docs, Copilot security: https://code.visualstudio.com/docs/copilot/security (read 12 Nov 2026) - Microsoft: VS Code docs, agent tools: https://code.visualstudio.com/docs/copilot/agents/agent-tools (read 12 Nov 2026) - Microsoft: VS Code docs, agent mode in chat: https://code.visualstudio.com/docs/copilot/chat/chat-agent-mode (read 12 Nov 2026) - Microsoft: VS Code docs, MCP servers: https://code.visualstudio.com/docs/copilot/customization/mcp-servers (read 12 Nov 2026) - Microsoft: VS Code docs, AI settings in enterprise environments: https://code.visualstudio.com/docs/enterprise/ai-settings (read 12 Nov 2026) - GitHub: Docs, Copilot cloud agent risks and mitigations: https://docs.github.com/en/copilot/concepts/agents/coding-agent/risks-and-mitigations (read 12 Nov 2026) - GitHub: Docs, customizing or disabling the agent firewall: https://docs.github.com/en/copilot/how-tos/use-copilot-agents/coding-agent/customize-the-agent-firewall (read 12 Nov 2026) - Microsoft: Defender AI agent support matrix: https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-support-matrix (read 12 Nov 2026) - Microsoft: Defender AI agent runtime protection overview: https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-runtime-protection-overview (read 12 Nov 2026) - CrowdStrike: Falcon Guardian blog post: https://www.crowdstrike.com/en-us/blog/crowdstrike-advances-ai-security-and-safety-falcon-guardian/ (read 12 Nov 2026) - Zenity: coding and personal agents use case: https://zenity.io/use-cases/agent-type/coding-personal-agents (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/guides/github-copilot-agent-mode-security/ --- # Codex CLI starts with the network off and writes limited to the workspace, so the work is keeping it that way GUIDE, read 12 Nov 2026 **The short answer.** OpenAI's Codex CLI runs commands in an OS-enforced sandbox with network off and writes limited to the workspace by default, plus an approval policy for leaving it. Three sandbox modes, requirements.toml for admins and OpenTelemetry export are covered, along with where the defaults stop and what remains. ## Codex separates what the sandbox allows from when the agent must ask OpenAI's Codex security documentation builds on two settings. The sandbox mode sets what Codex can do technically when it runs model-generated commands, for example where it can write and whether it can reach the network. The approval policy sets when Codex must stop and ask before it acts, for example before leaving the sandbox, using the network or running commands outside a trusted set. The pairing matters for audits. A laptop with a tight sandbox and a permissive approval policy behaves very differently from the reverse, and a config file review needs both values. The documentation says that for the Codex CLI and IDE extension, operating-system-level mechanisms enforce the sandbox policy, with defaults of no network access and write permissions limited to the active workspace. ## Three sandbox modes set the file and network boundary The configuration reference lists sandbox_mode values of read-only, workspace-write and danger-full-access. The default Auto preset pairs workspace-write with on-request approvals. In that preset, per OpenAI's documentation, Codex can read files, edit files and run commands in the working directory automatically, and it asks to edit files outside the workspace or to run commands that need network access. Folders under version control start in that Auto preset. Non-version-controlled folders start read-only, and Codex may also start read-only until the developer explicitly trusts the working directory. The reference also describes permission profiles through default_permissions, with built-in profiles named :read-only, :workspace and :danger-full-access, and custom profiles defined in [permissions.] tables. The reference says not to combine default_permissions with sandbox_mode. Approval prompts can be removed. The documentation says --ask-for-approval never, or -a never, disables them, and a --yolo flag is mentioned in the web search guidance. Treat any laptop configuration that sets danger-full-access or never as a finding to review rather than a preference. ## Network is off in workspace-write mode and is opened by one setting The documentation states that the default workspace-write sandbox keeps network access off unless the developer enables it in configuration. The switch is sandbox_workspace_write.network_access, which can also be passed on the command line with -c. That makes the setting easy to audit: search user and project configuration for it. Web search is controlled separately from command networking. Codex defaults to a web search cache, an OpenAI-maintained index of results, so cached mode returns pre-indexed results rather than fetching live pages. OpenAI says this reduces exposure to prompt injection from arbitrary live content and that web results should still be treated as untrusted. Where managed network rules are used, the docs describe destination rules that apply to scripts, programs and subprocesses spawned by commands, and a managed command allowlist for commands that use the managed proxy. OpenAI is clear on one limit: where policy permits full sandbox escalation and the escalation is approved, that execution can bypass the command proxy. Its advice is to configure enforced approval and sandbox requirements for the intended boundary and to test both ordinary and escalated commands. Two proxy options carry dangerous in their names: dangerously_allow_non_loopback_proxy can expose proxy listeners beyond the machine, and dangerously_allow_all_unix_sockets bypasses the Unix socket allowlist. ## Protected paths keep the agent away from repository and credential directories Inside writable roots, the default workspace-write policy still protects some paths. OpenAI documents that a .git directory (or the resolved directory when .git is a pointer file) is read-only, and so are .agents, .codex and .aws when they exist as directories. The list protects the places an agent could use to change hooks or configuration, or to alter cloud credentials. OpenAI's page tells teams to choose sandbox, approval and network settings by risk tolerance. The protected-path list names write restrictions. For read access, none found on 12 Nov 2026 in the pages read for this guide: a statement that credential files in a home directory are unreadable. Test what a session can read on your own build before you assume they are out of reach. ## Approvals, Auto-review and MCP tool annotations decide what interrupts the developer On-request approvals let commands the sandbox allows run without prompting and ask when Codex needs more. Codex can also route eligible approval requests to an automatic reviewer: set approvals_reviewer to auto_review, which applies when approvals are interactive, such as under on-request or a granular approval policy. The documentation's other example value is user. As with any model-based reviewer, treat it as a convenience that reduces prompts, and keep sandbox and network limits as the hard stop. Codex also elicits approval for app (connector) tool calls that advertise side effects, even when the action is not a shell command or file change. Per the documentation, destructive app or MCP tool calls always require approval when the tool advertises a destructive annotation, unless the tool advertises a read annotation, which takes priority. That rule depends on what the server claims about itself, so an unreviewed MCP server is a trust decision, not a control. For MCP servers, the configuration reference provides per-server enabled_tools and disabled_tools lists. The disabled list is applied after the enabled list, which makes it a usable deny mechanism for a tool you do not want a server to expose. ## Admin requirements constrain the settings a developer cannot be left to choose OpenAI's managed configuration documentation separates requirements from defaults. Requirements, set in requirements.toml, constrain security-sensitive settings: approval policy, approvals reviewer, Auto-review policy, sandbox mode, permission profiles, web search mode, managed hooks, which MCP servers users can enable and which plugin marketplace sources they can use. When a developer's value conflicts with an enforced rule, the client applies the rule. Omitted keys remain unconstrained. The older managed_config.toml supplies starting values that users can still change during a run, and the client reapplies them at the next start. Delivery is through the Agent security area of the admin console, local system requirements or macOS MDM. The precedence rule for local execution is that MDM and legacy managed-device requirements rank above Agent Security, which ranks above the device's system requirements file. OpenAI notes that its cloud-managed requirements do not apply to Work cloud containers, which keep their own policies, so test local and cloud controls separately. On hooks, the Codex repository's configuration page says admins can set allow_managed_hooks_only = true in requirements.toml to ignore user, project and session hook configurations while still allowing managed hooks. It is supported only in requirements.toml; putting it in config.toml does not enable the mode. The docs also warn that the retired approval policy value untrusted can stop clients from starting, so remove it from old profiles and startup scripts. ## OpenTelemetry export is available, and prompt logging is opt-in The configuration reference lists an [otel] section with an exporter set to none, otlp-http or otlp-grpc, an endpoint, headers and TLS certificate paths. Raw user prompts are exported only when otel.log_user_prompt is set to true. The reference also describes an analytics.enabled switch for the built-in metrics exporter. Set the exporter through managed configuration if you want command-level records from every laptop. Verify what your own export contains, command text included, before you rely on it for investigations. ## An endpoint control covers local scope and the approved escalation Codex's local defaults are tighter than most agents here, and the remaining risk concentrates in three places: a developer who changes the sandbox mode or network setting, an approved escalation outside the sandbox, and MCP servers whose behavior the annotations do not describe. Those are the cases where a second control that records and can stop commands adds something. Vendors name Codex in their coverage claims. Pillar says its runtime component sits in the pre-tool hook for Claude Code, Codex CLI and Cursor and decides on every action before it runs. CrowdStrike says hooks for OpenAI Codex can inspect prompts and block malicious activity before harmful actions run. Zenity says native hooks and its MCP gateway can block or modify a dangerous Codex tool call before it executes. Cato's AI Endpoint page shows Codex among example local agents with per-user risk flags. These are vendor claims, read on 12 Nov 2026. Check each one against the managed hooks setting above: an endpoint hook that a developer can disable by editing user configuration is a monitoring aid, while one delivered as a managed hook is harder to remove. ## Where this sits in the index - [Agent Surface Map](/agent-surface-map/): Which of six agent surfaces each of 16 tools sees - [Action Control Matrix](/action-control-matrix/): Which of six agent actions each tool can stop - [The ranked index](/ranking/): 16 tools on seven published criteria ## Related guides - [Securing Google Gemini CLI on developer laptops](/guides/gemini-cli-security/) - [What coding agents can touch on a developer laptop, and which controls ship with each](/guides/coding-agents-on-developer-laptops/) - [Securing GitHub Copilot agent mode and the cloud agent](/guides/github-copilot-agent-mode-security/) ## Sources - OpenAI: Codex docs, agent approvals and security: https://developers.openai.com/codex/agent-approvals-security (read 12 Nov 2026) - OpenAI: Codex docs, configuration reference: https://developers.openai.com/codex/config-reference (read 12 Nov 2026) - OpenAI: Codex docs, managed configuration: https://developers.openai.com/codex/enterprise/managed-configuration (read 12 Nov 2026) - OpenAI: github.com/openai/codex docs/config.md: https://raw.githubusercontent.com/openai/codex/main/docs/config.md (read 12 Nov 2026) - OpenAI: github.com/openai/codex docs/sandbox.md: https://raw.githubusercontent.com/openai/codex/main/docs/sandbox.md (read 12 Nov 2026) - Pillar Security: Agentic Endpoint platform page: https://www.pillar.security/platform/agentic-endpoint (read 12 Nov 2026) - CrowdStrike: Falcon Guardian blog post: https://www.crowdstrike.com/en-us/blog/crowdstrike-advances-ai-security-and-safety-falcon-guardian/ (read 12 Nov 2026) - Zenity: coding and personal agents use case: https://zenity.io/use-cases/agent-type/coding-personal-agents (read 12 Nov 2026) - Cato Networks: AI Endpoint platform page: https://www.catonetworks.com/platform/cato-ai-endpoint/ (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/guides/codex-cli-security/ --- # Gemini CLI needs the sandbox and folder trust switched on, because both ship off GUIDE, read 12 Nov 2026 **The short answer.** Google's Gemini CLI documents sandboxing through Seatbelt, Docker, Podman, gVisor and a Windows native option, but tools.sandbox defaults to undefined and Trusted Folders is disabled by default. System settings, a policy engine with admin tier, MCP allowlists and OpenTelemetry export cover the rest. ## Gemini CLI ships with the main protections opt-in Two of Gemini CLI's protective features start switched off. The configuration reference lists tools.sandbox with a default of undefined, meaning no sandbox unless the developer or an administrator sets one. The Trusted Folders documentation states that the feature is disabled by default and must be enabled in user settings with security.folderTrust.enabled. A third default sits in telemetry, which is also disabled until enabled. The practical result is that an unconfigured laptop runs Gemini CLI with confirmation prompts, an approval mode and nothing else. The rest of this guide covers what to switch on, and where the project's own documentation says the limits are. ## Sandboxing offers several backends and a network switch Google's sandboxing guide describes the benefit as isolating potentially dangerous operations such as shell commands and file modifications from the host. A developer enables it with the -s flag, the GEMINI_SANDBOX environment variable or the sandbox setting in settings.json. The documented backends are the macOS Seatbelt profiles through sandbox-exec, container-based sandboxing with Docker or Podman, a Windows native sandbox and gVisor (runsc) on Linux. On macOS, the guide lists six built-in Seatbelt profiles chosen with the SEATBELT_PROFILE variable: permissive-open, permissive-proxied, restrictive-open, restrictive-proxied, strict-open and strict-proxied. The default is permissive-open, which Google describes as denying operations by default, confining writes to the project directory and allowing broad file reads and network access. The strict profiles add read restrictions, and the proxied variants send network traffic through a proxy. For container sandboxes, Google documents a custom image through GEMINI_SANDBOX_IMAGE and a project-level .gemini/sandbox.Dockerfile, plus SANDBOX_FLAGS for extra container arguments. The configuration reference adds tools.sandboxAllowedPaths for extra accessible paths and tools.sandboxNetworkAccess, which defaults to false. A gVisor sandbox with --network=none gives the CLI no external network access at all, and Google notes it then cannot reach the Gemini API. - **Choose a strict profile on macOS** when the machine holds credentials, since the default profile allows broad reads. - **Use sandbox expansion deliberately.** When a sandboxed command fails on permissions, or is flagged as needing more (npm install is Google's example), the CLI shows a Sandbox Expansion Request. Approval extends permissions for that run only. - **Review SANDBOX_FLAGS in shell profiles.** Extra container flags change the isolation the sandbox provides. ## Trusted Folders stops a cloned project from changing the CLI's behavior With security.folderTrust.enabled set, Gemini CLI asks the first time it runs in a folder: trust the folder, trust the parent folder, or do not trust. Before the choice, a discovery phase lists commands, MCP servers, hooks, skills and setting overrides found in the project, and the dialog flags dangerous settings such as auto-approval or a disabled sandbox. Decisions are saved in ~/.gemini/trustedFolders.json. An untrusted folder runs in a restricted safe mode. Per the documentation, workspace settings are ignored, project .env files are not loaded, extension management is restricted, tool auto-acceptance is disabled, automatic memory loading is disabled, MCP servers do not connect and custom commands are not loaded. This is the direct control against a repository that ships its own configuration, hooks or MCP definitions. Headless runs cannot show the dialog. If the folder is untrusted and the feature is enabled, the CLI exits with FatalUntrustedWorkspaceError. The documented ways around this are the --skip-trust flag and the GEMINI_CLI_TRUST_WORKSPACE=true variable, which trust the workspace for the session. Search CI definitions and wrapper scripts for both, because they turn the control off for whatever the pipeline checks out. Google links separate trust guidance for GitHub Actions. ## Approval modes, YOLO and the policy engine decide what runs unprompted The configuration reference defines the default approval mode as default, which prompts for approval, with auto_edit auto-approving edit tools and plan as read-only. YOLO mode, which auto-approves all tool calls, can only be enabled from the command line with --yolo or --approval-mode=yolo. Two settings let an administrator remove it: security.disableYoloMode disables YOLO even when a flag requests it, and admin.secureModeEnabled disallows YOLO and the Always allow options. Both default to false, so they have to be set. Gemini CLI also has a policy engine. Its rules are TOML, each with a decision of allow, deny or ask_user, and ask_user is treated as deny in non-interactive mode. Policies sit in tiers: Default, Workspace, User and Admin, where Admin rules override the others. The Admin tier reads from a system directory, /etc/gemini-cli/policies on Linux, and an adminPolicyPaths setting can add paths. Google's documentation calls deny the recommended way to exclude tools. It also warns that the Workspace tier is currently limited, so use User or Admin policies for rules you depend on. The tool allowlist tools.core restricts the built-in tools to a named set. Google's enterprise guide calls explicit allowlisting the most secure approach because any tool not listed cannot be used. ## System settings give administrators the final word, with a documented caveat Google's enterprise guide describes a system defaults file and a system overrides file, merged with user and workspace settings. The overrides file has the final say. Locations are /etc/gemini-cli/settings.json on Linux, C:\ProgramData\gemini-cli\settings.json on Windows and /Library/Application Support/GeminiCli/settings.json on macOS, and the GEMINI_CLI_SYSTEM_SETTINGS_PATH variable changes the path. For MCP server definitions, precedence is System, then Workspace, then User, so a user cannot override a server already defined at system level, but can add servers with new names. Because the path can be redirected by an environment variable, the guide suggests a wrapper script on PATH that sets the variable to the corporate file before it runs the real binary. The guide's own warning is the most important sentence in it: these patterns should not be considered a foolproof security boundary, and they are designed to prevent accidental misuse and enforce policy, not to defend against a malicious actor with local administrative rights. Plan any laptop deployment with that scope in mind and file permissions on the settings file set accordingly. ## MCP servers get allowlists, a trust flag to avoid, and environment redaction Gemini CLI's MCP guide documents mcp.allowed and mcp.excluded lists of server names. Each server entry can set includeTools and excludeTools, and excludeTools takes precedence when a tool is in both. The setting that deserves an audit is trust: when true, it bypasses all tool call confirmations for that server. Search user and project settings for it. The guide also documents environment sanitization. By default, the CLI redacts sensitive environment variables from the base environment before launching a server, including variables matching patterns such as *TOKEN*, *SECRET* and *PASSWORD*, so AWS keys or GitHub tokens in the shell do not leak to a server by accident. A variable placed explicitly in the server's env block is trusted and not redacted, and Google advises using variable expansion rather than hard-coded secrets. Remote servers over SSE or streamable HTTP can use OAuth 2.0. ## OpenTelemetry export is opt-in, and prompt text is included unless you turn it off Google's telemetry guide lists telemetry.enabled as false by default, with a target of local or gcp, an OTLP endpoint and protocol, an outfile option and a useCollector option. The logPrompts setting defaults to true, so once telemetry is enabled, prompts are included unless it is set to false. Decide that before turning export on for a fleet. The gemini_cli.tool_call event records the function name and arguments, duration, success, the decision (accept, reject, auto_accept or modify), whether the tool was native or MCP, and the MCP server name. That is enough to answer which tools ran unprompted and through which server, provided the export is delivered to a store the developer cannot edit. The configuration reference separately lists a privacy.usageStatisticsEnabled setting for usage statistics. ## An endpoint control fills the gap where Gemini CLI has no enforced layer Gemini CLI has an admin policy tier and system settings, which is more than some tools, but Google states that these are not a boundary against a user with local administrative rights, and the sandbox and trust features are off until someone enables them. An endpoint control can verify that those switches are actually set on each machine and record tool calls apart from the CLI's own telemetry. Vendor coverage of Gemini CLI is thinner than for Claude Code or Cursor in the evidence read for this site. Microsoft's Defender support matrix lists Gemini CLI among assistants discovered with MCP discovery on Windows. Pillar says Gemini CLI is scanned for configuration but runtime blocking was not listed for it as of October 2026. Bay says its cloud workspace product names Gemini CLI among agents it finds across virtual machines, Kubernetes and hosted sessions. All three are vendor claims, read on 12 Nov 2026. If Gemini CLI is in your fleet, ask each vendor whether it observes tool calls or only the install. ## Where this sits in the index - [Agent Surface Map](/agent-surface-map/): Which of six agent surfaces each of 16 tools sees - [Action Control Matrix](/action-control-matrix/): Which of six agent actions each tool can stop - [The ranked index](/ranking/): 16 tools on seven published criteria ## Related guides - [What coding agents can touch on a developer laptop, and which controls ship with each](/guides/coding-agents-on-developer-laptops/) - [What agentless means on an endpoint](/guides/what-agentless-means-on-an-endpoint/) - [Securing OpenAI Codex CLI on developer laptops](/guides/codex-cli-security/) ## Sources - Google: github.com/google-gemini/gemini-cli docs/cli/sandbox.md: https://raw.githubusercontent.com/google-gemini/gemini-cli/main/docs/cli/sandbox.md (read 12 Nov 2026) - Google: github.com/google-gemini/gemini-cli docs/cli/trusted-folders.md: https://raw.githubusercontent.com/google-gemini/gemini-cli/main/docs/cli/trusted-folders.md (read 12 Nov 2026) - Google: github.com/google-gemini/gemini-cli docs/tools/mcp-server.md: https://raw.githubusercontent.com/google-gemini/gemini-cli/main/docs/tools/mcp-server.md (read 12 Nov 2026) - Google: github.com/google-gemini/gemini-cli docs/cli/enterprise.md: https://raw.githubusercontent.com/google-gemini/gemini-cli/main/docs/cli/enterprise.md (read 12 Nov 2026) - Google: github.com/google-gemini/gemini-cli docs/cli/telemetry.md: https://raw.githubusercontent.com/google-gemini/gemini-cli/main/docs/cli/telemetry.md (read 12 Nov 2026) - Google: github.com/google-gemini/gemini-cli docs/reference/policy-engine.md: https://raw.githubusercontent.com/google-gemini/gemini-cli/main/docs/reference/policy-engine.md (read 12 Nov 2026) - Google: github.com/google-gemini/gemini-cli docs/reference/configuration.md: https://raw.githubusercontent.com/google-gemini/gemini-cli/main/docs/reference/configuration.md (read 12 Nov 2026) - Microsoft: Defender AI agent support matrix: https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-support-matrix (read 12 Nov 2026) - Pillar Security: Agentic Endpoint platform page: https://www.pillar.security/platform/agentic-endpoint (read 12 Nov 2026) - Bay: cloud workspaces platform page: https://www.bay.io/platform/cloud-workspaces (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/guides/gemini-cli-security/ --- # Every coding agent acts as the developer, and their defaults differ more than their feature lists GUIDE, read 12 Nov 2026 **The short answer.** Claude Code, Cursor, GitHub Copilot agent mode, Codex CLI and Gemini CLI all run shell commands as the signed-in developer. Their own docs differ on sandbox defaults, network defaults and managed policy. The guide compares files, shell, network, credentials and packages, then lists the gaps an endpoint control can fill. ## All five agents run commands as the developer, so the user account is the real boundary Claude Code, Cursor, GitHub Copilot agent mode in VS Code, OpenAI Codex CLI and Google Gemini CLI share one basic property. Each can read files, edit them and run shell commands on a developer laptop using the signed-in developer's account. Anthropic states the consequence for its own tool: a Bash command the developer approves can still write anywhere the user account can. Cursor describes its Run Modes as best-effort guardrails rather than a hard security boundary. Google says its enterprise settings are not a foolproof boundary against a user with local administrative rights. Each vendor's controls therefore reduce accidents and limit blast radius. They differ in what is on at install, what an administrator can force, and what leaves a record. This guide compares the five using each tool's own security documentation, read on 12 Nov 2026. The per-tool guides on this site go deeper on configuration. ## Files, shell, network, credentials and packages are the five surfaces to compare The table lists what each agent can reach by default, or by documented option, in each surface. Cells state the default behavior from the docs, and a setting name where one decides the outcome. | Surface | Claude Code | Cursor | Copilot agent mode | Codex CLI | Gemini CLI | | --- | --- | --- | --- | --- | --- | | Files | Manual mode asks before edits; sandbox, if enabled, reads most of the machine unless denyRead is set | Reads and workspace edits need no approval; config files do; Read Access defaults to System | Edits tracked with Keep and Undo; autoApprove globs protect sensitive files | workspace-write: writes limited to workspace; .git, .codex, .aws protected | Seatbelt default profile confines writes to the project and allows broad reads | | Shell | Approval per command in Manual mode; sandbox optional | Approval by default; sandbox in Auto-review and optional in Allowlist | Approval by default; per-command auto-approve rules; sandbox on supported platforms | Runs in OS sandbox; asks to leave it | Confirmation prompts; sandbox off unless set | | Network | Sandbox proxy with allowedDomains starting empty, if sandbox is on | Sandbox blocks by default; agent tools limited to listed destinations | Sandbox allowNetwork defaults to true | Off by default in workspace-write | sandboxNetworkAccess setting defaults to false; the default Seatbelt profile allows network | | Credentials | Stored in Keychain on macOS; ~/.ssh readable in sandbox unless denied | Protected paths include .git/config and Cursor config; Allowed Team IDs blocks personal accounts | MCP secrets in VS Code secure store | .aws and .codex protected as read-only | MCP servers get environment variables redacted by default | | Packages | None found on 12 Nov 2026 in the pages read | None found on 12 Nov 2026 in the pages read | Cloud agent firewall allowlist permits dependency downloads | Legacy cloud: setup phase installs dependencies; agent phase offline by default | Sandbox expansion flags npm install as needing extra permission | ## Each tool ships a different set of administrator controls The second table covers what an administrator can enforce and what telemetry exists. A control counts here only when the vendor documents it as enforceable by the organization. | Control | Claude Code | Cursor | Copilot | Codex CLI | Gemini CLI | | --- | --- | --- | --- | --- | --- | | Managed policy | managed-settings.json, MDM, server-managed settings | Team dashboard policies; MDM for workspace trust | Enterprise-managed settings; VS Code device policy | requirements.toml via admin console, system file or MDM | System settings file; Admin policy tier | | Disable the unsafe mode | disableBypassPermissionsMode, disableAutoMode | Team Run Modes setting exists; a switch that disables Run Everything: none found on 12 Nov 2026 in the pages read | Managed permission and sandbox settings exist (key names not read) | allowed approval policies and sandbox constraints | security.disableYoloMode, admin.secureModeEnabled | | MCP restriction | allowedMcpServers, deniedMcpServers, managed-mcp.json | Per-call approval; MCP allowlist; team marketplace | allowedMcpServers, deniedMcpServers | Managed list of enabled servers; per-tool allow and deny | mcp.allowed, mcp.excluded, system-level definitions | | Pre-execution hook | PreToolUse can block; allowManagedHooksOnly | beforeShellExecution, beforeMCPExecution | None found on 12 Nov 2026 in the pages read | Managed hooks; allow_managed_hooks_only | Policy engine allow, deny, ask_user | | Telemetry | OTel, opt-in; tool_decision and tool_result events | Audit log covers settings, not agent activity; hooks for activity | None found on 12 Nov 2026 in the pages read | OTel exporter; prompt logging opt-in | OTel off by default; tool_call events; logPrompts defaults true | ## Defaults differ more than the feature lists suggest Read in the order of install, the five agents start in different places. Codex CLI starts with an OS-enforced sandbox, network off and writes limited to the workspace. Claude Code starts in a permission mode with prompts and an optional sandbox, off by default. Gemini CLI also leaves its sandbox and Trusted Folders off. Cursor asks before shell commands and MCP calls but lets the agent read widely. Copilot agent mode asks before terminal commands and has a sandbox option whose network setting defaults to open. Defaults are what an unmanaged laptop runs. A team that publishes a hardened configuration is judging the tool at its managed state, and one that does not is judging it at install. Check both before comparing. ## Repository files, MCP servers and bypass modes are the shared weak points Three patterns recur across the documentation. First, a repository can carry configuration that changes agent behavior. Claude Code asks for workspace trust and per-server approval for .mcp.json; Gemini CLI's Trusted Folders discovery lists commands, MCP servers, hooks, skills and setting overrides; VS Code applies Workspace Trust to MCP definitions; Cursor's workspace trust is off by default. Anthropic adds that a non-interactive -p session shows neither the trust dialog nor the server prompt, which matters for scripts. Second, MCP servers run with the developer's permissions and no vendor audits them. Anthropic says it does not security-audit or manage any MCP server, VS Code says local servers can run code on the machine, and Codex relies on tool annotations that the server itself advertises. Third, every tool has a mode that removes the prompts: bypassPermissions in Claude Code, Run Everything in Cursor, Allow all and Autopilot in Copilot, never and danger-full-access in Codex, and YOLO in Gemini CLI. Only some have a documented administrator switch to disable it, as the second table shows. A fourth gap is common to the whole set. In the pages read, none found on 12 Nov 2026 describes a control that checks the reputation or integrity of packages an agent installs. The sandbox and firewall features decide whether a download can happen, not whether it is safe. ## An endpoint control adds pre-execution blocking and a record the agent cannot edit The agent's own settings can be edited by a user with the right access, and the telemetry they emit is either off by default or limited. An endpoint control adds two things the tools do not provide uniformly. The first is a decision before a command or tool call runs, applied the same way across agents. The second is a log delivered off the machine. Several of the tools document the interface vendors use for the first: Claude Code PreToolUse hooks, Cursor beforeShellExecution and beforeMCPExecution, and Codex managed hooks. Vendors state their coverage as follows, and each statement is the vendor's claim, read on 12 Nov 2026. CrowdStrike says agent-level hooks for Claude Code, OpenAI Codex, GitHub Copilot, OpenCode and Cursor can inspect prompts and block malicious activity before harmful actions run, and that its sensor blocks malicious open-source npm and PyPI packages as they reach the endpoint, including packages pulled in by AI coding agents. Pillar says pre-action enforcement covers Claude Code, Codex CLI and Cursor only as of October 2026, with Gemini CLI scanned for configuration but runtime blocking not listed. Microsoft says agent-native inspection uses vendor event interfaces for Claude Code, Codex CLI and GitHub Copilot CLI. Zenity says it can block or modify a dangerous tool call before it executes for Claude Code, Codex, Copilot and Cursor. Noma says it uses Cursor Agent Hooks for inline decisions. The package claim is the only one in this set that addresses the fourth gap. The coverage differs by agent, which is why a comparison table from a vendor deserves a test on your own fleet: install each of the five agents, trigger a blocked action and confirm what the tool recorded. ## A one-afternoon check on a single laptop shows which gaps are open Before choosing any tool, a security engineer can test one reference laptop. This is verification of settings, not a rollout program. Each check below comes from a setting named in the vendors' documentation. - **Sandbox state.** Claude Code: sandbox.enabled and failIfUnavailable. Gemini CLI: tools.sandbox. Copilot: agent sandboxing and the allowNetwork setting. Codex: sandbox_mode and sandbox_workspace_write.network_access. - **Prompt-removal modes.** Confirm whether the disable switches are set: Claude Code's disableBypassPermissionsMode, Gemini CLI's disableYoloMode, and Codex's approval and sandbox requirements. - **Repository trust.** Check Trusted Folders in Gemini CLI, Workspace Trust in VS Code and Cursor, and search CI scripts for Gemini CLI's trust bypass flag. - **Telemetry destination.** Confirm OpenTelemetry settings are in managed configuration, and decide on prompt logging: Gemini CLI's logPrompts defaults to true and Codex's log_user_prompt is opt-in. - **Credential reach.** Confirm which paths a sandboxed command can read, starting with ~/.ssh and ~/.aws. ## Where this sits in the index - [Agent Surface Map](/agent-surface-map/): Which of six agent surfaces each of 16 tools sees - [Action Control Matrix](/action-control-matrix/): Which of six agent actions each tool can stop - [The ranked index](/ranking/): 16 tools on seven published criteria ## Related guides - [What agentless means on an endpoint](/guides/what-agentless-means-on-an-endpoint/) - [Local MCP servers on the endpoint](/guides/local-mcp-servers-on-the-endpoint/) - [Securing Google Gemini CLI on developer laptops](/guides/gemini-cli-security/) ## Sources - Anthropic: Claude Code docs, security: https://code.claude.com/docs/en/security (read 12 Nov 2026) - Anthropic: Claude Code docs, sandboxing: https://code.claude.com/docs/en/sandboxing (read 12 Nov 2026) - Anthropic: Claude Code docs, permissions: https://code.claude.com/docs/en/permissions (read 12 Nov 2026) - Anthropic: Claude Code docs, settings: https://code.claude.com/docs/en/settings (read 12 Nov 2026) - Anthropic: Claude Code docs, MCP: https://code.claude.com/docs/en/mcp (read 12 Nov 2026) - Anthropic: Claude Code docs, hooks: https://code.claude.com/docs/en/hooks (read 12 Nov 2026) - Anthropic: Claude Code docs, monitoring usage: https://code.claude.com/docs/en/monitoring-usage (read 12 Nov 2026) - Cursor: docs, agent security: https://cursor.com/docs/agent/security (read 12 Nov 2026) - Cursor: docs, run modes and sandboxing: https://cursor.com/docs/agent/security/run-modes (read 12 Nov 2026) - Cursor: docs, privacy and data governance: https://cursor.com/docs/enterprise/privacy-and-data-governance (read 12 Nov 2026) - Cursor: docs, compliance and monitoring: https://cursor.com/docs/enterprise/compliance-and-monitoring (read 12 Nov 2026) - Cursor: docs, hooks: https://cursor.com/docs/hooks (read 12 Nov 2026) - Microsoft: VS Code docs, Copilot security: https://code.visualstudio.com/docs/copilot/security (read 12 Nov 2026) - Microsoft: VS Code docs, MCP servers: https://code.visualstudio.com/docs/copilot/customization/mcp-servers (read 12 Nov 2026) - Microsoft: VS Code docs, AI settings in enterprise environments: https://code.visualstudio.com/docs/enterprise/ai-settings (read 12 Nov 2026) - GitHub: Docs, customizing or disabling the agent firewall: https://docs.github.com/en/copilot/how-tos/use-copilot-agents/coding-agent/customize-the-agent-firewall (read 12 Nov 2026) - OpenAI: Codex docs, agent approvals and security: https://developers.openai.com/codex/agent-approvals-security (read 12 Nov 2026) - OpenAI: Codex docs, configuration reference: https://developers.openai.com/codex/config-reference (read 12 Nov 2026) - OpenAI: Codex docs, managed configuration: https://developers.openai.com/codex/enterprise/managed-configuration (read 12 Nov 2026) - Google: github.com/google-gemini/gemini-cli docs/cli/sandbox.md: https://raw.githubusercontent.com/google-gemini/gemini-cli/main/docs/cli/sandbox.md (read 12 Nov 2026) - Google: github.com/google-gemini/gemini-cli docs/cli/trusted-folders.md: https://raw.githubusercontent.com/google-gemini/gemini-cli/main/docs/cli/trusted-folders.md (read 12 Nov 2026) - Google: github.com/google-gemini/gemini-cli docs/tools/mcp-server.md: https://raw.githubusercontent.com/google-gemini/gemini-cli/main/docs/tools/mcp-server.md (read 12 Nov 2026) - Google: github.com/google-gemini/gemini-cli docs/cli/enterprise.md: https://raw.githubusercontent.com/google-gemini/gemini-cli/main/docs/cli/enterprise.md (read 12 Nov 2026) - Google: github.com/google-gemini/gemini-cli docs/cli/telemetry.md: https://raw.githubusercontent.com/google-gemini/gemini-cli/main/docs/cli/telemetry.md (read 12 Nov 2026) - Google: github.com/google-gemini/gemini-cli docs/reference/configuration.md: https://raw.githubusercontent.com/google-gemini/gemini-cli/main/docs/reference/configuration.md (read 12 Nov 2026) - CrowdStrike: Falcon Guardian blog post: https://www.crowdstrike.com/en-us/blog/crowdstrike-advances-ai-security-and-safety-falcon-guardian/ (read 12 Nov 2026) - CrowdStrike: supply chain attacks endpoint blog post: https://www.crowdstrike.com/en-us/blog/crowdstrike-extends-endpoint-security-to-stop-supply-chain-attacks/ (read 12 Nov 2026) - Pillar Security: Agentic Endpoint platform page: https://www.pillar.security/platform/agentic-endpoint (read 12 Nov 2026) - Microsoft: Defender AI agent runtime protection overview: https://learn.microsoft.com/en-us/defender-endpoint/ai-agent-runtime-protection-overview (read 12 Nov 2026) - Zenity: coding and personal agents use case: https://zenity.io/use-cases/agent-type/coding-personal-agents (read 12 Nov 2026) - Noma Security: runtime security for Cursor blog post: https://noma.security/blog/securing-the-agentic-frontier-noma-unveils-the-first-real-time-agent-runtime-security-for-cursor (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/guides/coding-agents-on-developer-laptops/ --- # Agentless on an endpoint describes how a tool is delivered, and each delivery method sees a different slice of AI activity GUIDE, read 12 Nov 2026 **The short answer.** Agentless AI endpoint security can mean an MDM-pushed profile, a browser extension, an OS API sensor, existing EDR telemetry or a SaaS API connector. Each sees a different slice. Browser extensions miss terminal agents, and API connectors miss laptops. Test any claim by running one coding agent, one desktop app and one browser agent. ## Agentless is a statement about install method, not about what a tool can see When a vendor says a product is agentless, the claim usually means one thing: no new endpoint sensor ships to the laptop. It does not say where the product gets its data. A tool that reads configuration files through your MDM, a tool that lives inside the browser, a tool that calls a SaaS vendor's admin API and a tool that reads the telemetry your EDR already collects are all described as agentless. They share a label and almost nothing else. For AI endpoint security the difference matters more than it did for classic endpoint tools. The activity you care about is spread across at least six surfaces: desktop AI apps, IDE assistants, command-line agents, MCP servers, browser extensions and SaaS copilots. No single delivery method reaches all six. A browser extension cannot see a terminal agent. An API connector cannot see a process on a laptop. EDR telemetry can see the process but usually not the prompt or the tool call inside it. This guide lists the five delivery methods that get called agentless, says what each can and cannot see, and ends with a test you can run against a vendor in an afternoon. Vendor product claims are written as the vendor's. This page makes no claim about any specific product's coverage. ## MDM-pushed profiles configure the agents that already run, but they do not watch them A mobile device management (MDM) profile is a settings payload the operating system applies. Microsoft Intune, for example, lets you add custom settings to macOS devices by supplying an .xml or .mobileconfig file, per Microsoft's Intune documentation (read 12 Nov 2026). Windows uses registry policy or Group Policy for the same job. This is a real control channel for AI tools. Anthropic documents that Claude Code accepts endpoint-managed settings deployed through native OS policies (macOS managed preferences, Windows registry) or a managed settings file, and that these sit above user settings in precedence (Anthropic, Claude Code server-managed settings page, read 12 Nov 2026). Claude Code also reads a managed-mcp.json from a fixed path per platform: /Library/Application Support/ClaudeCode on macOS, /etc/claude-code on Linux and WSL, and C:\Program Files\ClaudeCode on Windows (Anthropic, Claude Code managed MCP page, read 12 Nov 2026). VS Code publishes enterprise policies for MCP access and an MCP server allowlist (Microsoft, VS Code enterprise policies page, read 12 Nov 2026). Chrome and Edge take extension allow and block lists through policy. What a profile cannot do is tell you what happened. It sets a rule and the application enforces it, or does not. If a user runs an AI tool that reads no managed policy, the profile has nothing to say about it. A profile is a good way to turn off what you have decided to turn off, and a poor way to learn what you have not yet found. - Sees: nothing by itself. It is a configuration channel, not a sensor. - Can enforce: allowlists and denylists for tools that read managed settings, extension install rules, MCP server restrictions. - Misses: any AI tool that ignores managed settings, any tool you did not know to write a policy for, and any action taken inside an allowed tool. ## Browser extensions see the page and the clipboard, and stop at the browser edge An extension deployed to Chrome or Edge by policy runs inside the browser with the permissions its manifest requests. Google documents that permission warnings describe the capabilities an API grants, and that a host permission such as all URLs lets an extension act on every site (Chrome for Developers, permission warnings page, read 12 Nov 2026). That is why a security extension can see prompts typed into a web chat, files chosen in an upload dialog and text pasted into a page. Microsoft's own endpoint DLP shows the pattern. After you install the Microsoft Purview extension for Chrome on Windows 10 or 11, the organization can monitor attempts to access or upload a sensitive item to a cloud service in Chrome and enforce protective actions (Microsoft Learn, Purview extension for Chrome page, read 12 Nov 2026). The edge of this approach is the browser itself. A coding agent in a terminal, a desktop AI app built on its own runtime, and a local MCP server make network calls and file reads without touching a page. Another browser, an Incognito window where extensions are not allowed, or a browser profile your policy does not cover will also fall outside it. Google's Chrome Enterprise help describes Incognito as a way users can escape extension-based controls unless you manage it with the IncognitoModeAvailability policy (Google, Chrome Enterprise Help, read 12 Nov 2026). - Sees: prompts, pastes, uploads and page content in the managed browser. - Can enforce: block, warn or redact at the moment of paste or upload, in the browser. - Misses: terminal agents, desktop AI apps, local MCP servers, other browsers, unmanaged profiles. ## OS APIs give process and file events, but they need code running on the machine Operating systems expose security events to approved software. On macOS that route is Apple's Endpoint Security framework, documented on Apple's developer site (read 12 Nov 2026). It is how a security product learns that a process started, a file was opened or an executable ran. Windows has equivalent event sources. These APIs are the strongest source for the questions an AI endpoint team asks first: which AI binaries run on which machines, which child processes did a coding agent start, which files did it open. They do not carry the meaning of the action. A shell command started by a coding agent looks, at the process level, like a shell command started by a developer. Telling them apart means linking the process to its parent, to the application that launched it, and ideally to the agent session. A reader of vendor material should notice one point. Anything that consumes these APIs is code on the machine, whatever the marketing page calls it. That is not a flaw. It is the price of seeing process and file events. If a vendor says agentless and also says it sees every command an agent runs, ask which component on the laptop produces that data. - Sees: process start, file access, network connections, depending on the sensor. - Can enforce: block a process or file operation, if the sensor runs in an enforcing mode. - Misses: the prompt and the agent's reason for acting, unless the product also reads the agent's own logs or hooks. ## Existing EDR telemetry is reuse, and its depth depends on what your EDR already collects Some vendors position themselves as agentless because they read data from the EDR you already pay for. That is a legitimate design, and it keeps the deployment small. The limit is that the product can only see what the EDR was configured to record. If your EDR keeps process starts for 30 days but does not record command-line arguments for some process classes, the AI tool inherits that gap. Ask what the integration pulls, how often, and what it does with it. A read-only feed supports inventory and detection after the fact. It supports blocking only if the product can also send a response action back to the EDR, and that response then depends on the EDR's own latency. Whether a given EDR is the right base for AI coverage is a separate question. Here the point is narrower: reuse of EDR telemetry is a delivery method with its own boundary, and the boundary is the EDR's data model. - Sees: processes, command lines, network connections and file events your EDR already records. - Can enforce: only what the EDR's response actions allow. - Misses: anything the EDR does not record, and the content of prompts and tool calls. ## API connectors reach SaaS copilots and OAuth grants, not the laptop An API connector authenticates to a SaaS service with an admin credential and reads what the service exposes. For AI, the useful data is the list of third-party apps that users have connected to a tenant. Microsoft documents how to review permissions granted to applications in a Microsoft Entra tenant, covering applications added through user or admin consent (Microsoft Learn, manage application permissions page, read 12 Nov 2026). Similar admin views exist in other identity and productivity platforms. This is the cleanest agentless method, and it is entirely accurate to call it agentless. It shows which AI apps hold OAuth access to mail, files and calendars, and which scopes they hold. It does not show a single thing about a laptop. A local MCP server, a command-line agent that uses a personal API key, and a desktop app that never touched your identity provider all stay invisible to it. Treat an API connector as a source for the SaaS copilot surface and for OAuth grants, and judge any broader claim against that boundary. - Sees: connected apps, scopes, consent events, SaaS copilot usage where the vendor exposes it. - Can enforce: revoke a grant, disable an app, change a tenant setting. - Misses: processes, files, local configs and traffic on the device. ## A side-by-side view of what each method sees The table summarizes the five methods against the six surfaces this site maps. 'Yes' means the method can observe the surface in principle. It does not say a given product does. | Delivery method | Desktop AI apps | IDE assistants | Command-line agents | MCP servers | Browser extensions and AI browsers | SaaS copilots | | --- | --- | --- | --- | --- | --- | --- | | MDM profile | Configure only | Configure only | Configure only | Configure only | Configure only | No | | Browser extension | Web versions only | No | No | No | Yes | Web access only | | OS API sensor | Process and file events | Process and file events | Process and file events | Process and file events | Process events | No | | EDR telemetry reuse | Process events | Process events | Process events | Process events | Process events | No | | API connector | No | No | No | No | No | Yes | ## A four-step test for any agentless claim Ask the vendor for a trial on a small set of devices, and run the same four steps against every candidate. The steps use ordinary tools and need no attack content. Step one: name the data source. For each surface the vendor says it covers, get the exact component that produces the data: an MDM payload, an extension ID, a system extension, an EDR API, a SaaS API. If the answer is a laptop component, the product has an endpoint footprint, and you should plan for it. If the answer is an API, the product cannot see the laptop. Step two: run one tool per surface. Start a desktop AI app, a coding assistant in an IDE, a command-line agent, a local MCP server defined in a config file, an AI browser extension, and a SaaS copilot sign-in. Check which of these appear in the product's inventory within a day. Step three: make each tool do something observable and harmless. Have the coding agent read a file in a test folder, run a benign shell command and make one outbound request. Have the desktop app upload a test file. Compare what the product recorded with what you did, including the tool call name, the file path and the user. Step four: ask what happens when the product is turned off or removed. For an extension, can a user disable it? For a profile, can a user remove it? For a sensor, is there tamper protection? A control that the person under control can remove is a log, not a control. Record the results in a table like the one above and score what you saw, not what the data sheet says. The scoring approach used on this site follows the same rule: a vendor's own claim is capped at 7 of 10 on a criterion until a second source confirms it, and a feature with no evidence found scores 0. ## Where this sits in the index - [Agent Surface Map](/agent-surface-map/): Which of six agent surfaces each of 16 tools sees - [Action Control Matrix](/action-control-matrix/): Which of six agent actions each tool can stop - [The ranked index](/ranking/): 16 tools on seven published criteria ## Related guides - [Local MCP servers on the endpoint](/guides/local-mcp-servers-on-the-endpoint/) - [Browser agents and AI extensions](/guides/browser-agents-and-ai-extensions/) - [What coding agents can touch on a developer laptop, and which controls ship with each](/guides/coding-agents-on-developer-laptops/) ## Sources - Microsoft Learn: Intune custom settings for macOS: https://learn.microsoft.com/en-us/mem/intune/configuration/custom-settings-macos (read 12 Nov 2026) - Anthropic: Claude Code server-managed settings: https://code.claude.com/docs/en/server-managed-settings (read 12 Nov 2026) - Anthropic: Claude Code managed MCP: https://code.claude.com/docs/en/managed-mcp (read 12 Nov 2026) - Microsoft: VS Code enterprise policies: https://code.visualstudio.com/docs/enterprise/policies (read 12 Nov 2026) - Google: Chrome extension permission warnings: https://developer.chrome.com/docs/extensions/develop/concepts/permission-warnings (read 12 Nov 2026) - Microsoft Learn: Purview extension for Chrome: https://learn.microsoft.com/en-us/purview/dlp-chrome-learn-about (read 12 Nov 2026) - Google: Chrome Enterprise Help, extensions in Incognito: https://support.google.com/chrome/a/answer/13130396 (read 12 Nov 2026) - Apple: Endpoint Security framework: https://developer.apple.com/documentation/endpointsecurity (read 12 Nov 2026) - Microsoft Learn: manage application permissions: https://learn.microsoft.com/en-us/entra/identity/enterprise-apps/manage-application-permissions (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/guides/what-agentless-means-on-an-endpoint/ --- # Local MCP servers run with the user's full privileges, so find the config files first and restrict which servers may start GUIDE, read 12 Nov 2026 **The short answer.** A local MCP server is a process an AI client starts on a laptop, usually over stdio, with the user's own file and network rights. The MCP specification's security guidance names startup-command injection and unsandboxed execution as risks. Read the config files for Claude Desktop, Cursor, VS Code and Claude Code, then restrict by policy. ## A local MCP server is a program the AI client launches as the user The Model Context Protocol (MCP) lets an AI application call tools and read data through a standard interface. It defines two standard transports: stdio, where the client launches the server as a subprocess and talks to it over standard input and output, and Streamable HTTP (Model Context Protocol specification, transports page, read 12 Nov 2026). On a laptop, stdio is the common local case. The spec says clients SHOULD support stdio whenever possible. That design has a direct endpoint consequence. When a user adds a stdio server, the client config holds a command, its arguments and environment variables. The client runs that command. The resulting process has the same file, network and credential access as the user. The MCP security best practices page says this plainly: local MCP servers may have direct access to the user's system and may be accessible to other processes on the machine, which makes them attractive targets (modelcontextprotocol.io, security best practices page, read 12 Nov 2026). This guide covers servers on the laptop. Remote MCP servers reached over HTTP raise different questions, and gateways that sit in front of them are a separate product category that this site does not cover. ## Each client keeps its server list in a different file, so discovery needs one path per client There is no registry of which MCP servers a machine runs. The only inventory is the set of client config files, plus the processes those clients start. These are the locations documented by each vendor, read on 12 Nov 2026. | Client | Where the config lives | Source | | --- | --- | --- | | Claude Desktop | macOS: ~/Library/Application Support/Claude/claude_desktop_config.json. Windows: %APPDATA%\Claude\claude_desktop_config.json. Linux: ~/.config/Claude/claude_desktop_config.json | modelcontextprotocol.io, connect local servers | | Cursor | Project: .cursor/mcp.json. Global: ~/.cursor/mcp.json. Fields for stdio servers include type, command, args and env | Cursor docs, MCP | | VS Code | mcp.json files that VS Code reads for workspace and user servers, with enterprise policy keys ChatMCP (chat.mcp.access) and ChatAllowedMcpServers | VS Code docs, MCP servers and enterprise policies | | Claude Code | Project: .mcp.json in the repository. User and local scope: ~/.claude.json. Organization: managed-mcp.json at a fixed system path | Anthropic, Claude Code MCP and managed MCP | ## Project-level configs travel with the repository, which changes who can add a server Two of the four clients accept a config file inside the project folder: Cursor reads .cursor/mcp.json, and Claude Code reads .mcp.json in the repository. A project file can be committed to source control, which means a teammate or a cloned repository can introduce a server definition to a developer's machine. The user adds nothing by hand. Anthropic's documentation addresses this. Claude Code has a project server approval step and a workspace trust step that govern how approvals committed to a repository are treated (Anthropic, Claude Code MCP page, read 12 Nov 2026). Treat those prompts as part of your control surface. A developer who clicks through an approval for a repository they just cloned has given that repository's author a way to start a process on the laptop. This is the same pattern the MCP security page lists first among local-server attacks: an attacker includes a malicious startup command in a client configuration. The page names three attacks in all, which are a malicious startup command in a config, a malicious payload inside the server itself, and access to an insecure local server left running on localhost, through DNS rebinding. ## The MCP specification names the risks of local servers and the client controls that answer them The security best practices page has a section on local MCP server compromise. It says that without proper sandboxing and consent requirements, an attacker can include a malicious startup command in a client configuration, distribute a malicious payload inside the server, or reach an insecure server left running on localhost through DNS rebinding. It lists the impacts as remote code execution with user privileges, session hijacking, credential theft and unauthorized actions (modelcontextprotocol.io, security best practices page, read 12 Nov 2026). Its mitigations are written for client authors, but each one maps to something an endpoint team can check or enforce: - Consent before connecting. For one-click configuration, the client MUST show the exact command, with arguments and parameters and without truncation, identify it as a potentially dangerous operation that executes code, and require explicit user approval. Check whether each client you allow does this. - Warn on dangerous patterns. The page asks clients to highlight commands that access sensitive locations such as the home directory, SSH keys and system directories. - Warn that servers run with the client's privileges. Users should be told that a server acts as them. - Sandbox the server. The page recommends running server commands in a sandboxed environment with minimal default privileges, restricted file system and network access, and a way for users to grant specific extra access. - Use platform sandboxing. Containers, chroot and application sandboxes are named as options, kept up to date. ## Inventory starts with the config files and ends with the running processes Discovery has two layers, and you need both. The config layer tells you which servers a user has defined. The process layer tells you which are running and what they started. For the config layer, collect the files in the table above from every user profile on the machine, not only the logged-in user. Parse out the command, args, env keys and url fields. Record the file path, a hash of the file and the modification time so that you can detect change. Do not collect the values of env entries: they often hold API tokens, and your inventory should not become a second copy of them. Pay attention to the command field. A server defined as npx, uvx, docker run or a bare interpreter pulls and executes code from a package registry or image at launch. A server defined as a path to a binary in the user's home directory is a program you have not reviewed. Both patterns are normal for MCP, and both are worth a line in the inventory. For the process layer, look for child processes of the AI clients: Claude Desktop, Cursor, VS Code and its derivatives, and command-line agents. A stdio server is a child of the client, so process ancestry separates an MCP server from an unrelated Node or Python process. Your EDR or an OS-level sensor can supply this ancestry. Record the parent, the executable path, the command line and any listening ports. ## Control comes from managed policy at the client, and each client offers a different lever Whichever route you choose, write the rule down as a list of approved servers by name and version, not as a general instruction to be careful. - Claude Code: deploy a managed-mcp.json to /Library/Application Support/ClaudeCode on macOS, /etc/claude-code on Linux and WSL, or C:\Program Files\ClaudeCode on Windows. Anthropic describes this as exclusive control: users get the fixed set you ship. Allow and deny lists (allowedMcpServers and deniedMcpServers) are available when you want users to keep their own servers inside limits. Deploy through MDM, Group Policy or Intune (Anthropic, Claude Code managed MCP page, read 12 Nov 2026). - VS Code: the ChatMCP policy controls access to installed MCP servers. ChatAllowedMcpServers is an allowlist, and when set, only matching servers may be installed or run. A deny list takes precedence over an allow list. McpGalleryServiceUrl sets which gallery the editor uses (Microsoft, VS Code enterprise policies page, read 12 Nov 2026). - Cursor and Claude Desktop: the documentation pages read for this guide describe config files and fields. None found on 12 Nov 2026 for an equivalent allowlist policy on those pages. If you need to restrict them, ask the vendor, and fall back to file-integrity monitoring on the config paths and to blocking the launch commands at the endpoint. ## A short checklist for a laptop fleet Run these steps in order. Each is a task, not a product feature. - Find every config file for the four clients on a sample of 20 machines and count servers per machine. The number sets the size of the problem. - Group servers by launch method: package runner, container, local binary, remote URL. Package runners and local binaries deserve first review. - Choose an approved list. Start from the servers your developers already use and that you can name an owner for. - Deploy managed policy where the client supports it, and set a monitored baseline where it does not. - Alert on a new or changed server definition, and on a process whose parent is an AI client and whose executable lives under a user-writable directory. - Review how each client displays the startup command at approval time, and train people to read it. - Reduce server privileges where you can: restrict the folders a filesystem server may read, and keep tokens out of env entries where a secret store is available. ## What vendors can add, and what to ask them A product that covers MCP on the endpoint should be able to show you, per device, the servers defined and the servers running, with the command that started each. It should show tool calls made through the server where it can read them, and it should be able to block a server from starting or a tool call from completing. Whether it can do the last of these before the action happens is the question that separates monitoring from control. Ask for three things in a trial: the config paths it reads, the process it watches for a stdio server's launch, and a demonstration of a blocked server on one of the four clients above. Treat what the vendor says as the vendor's claim until you have seen the block happen on your own machine. On this site's scorecard, an unconfirmed vendor claim is capped at 7 of 10. ## Where this sits in the index - [Agent Surface Map](/agent-surface-map/): Which of six agent surfaces each of 16 tools sees - [Action Control Matrix](/action-control-matrix/): Which of six agent actions each tool can stop - [The ranked index](/ranking/): 16 tools on seven published criteria ## Related guides - [Browser agents and AI extensions](/guides/browser-agents-and-ai-extensions/) - [AI DLP on the endpoint](/guides/ai-dlp-on-the-endpoint/) - [What agentless means on an endpoint](/guides/what-agentless-means-on-an-endpoint/) ## Sources - modelcontextprotocol.io: security best practices: https://modelcontextprotocol.io/docs/tutorials/security/security_best_practices (read 12 Nov 2026) - modelcontextprotocol.io: transports specification 2025-11-25: https://modelcontextprotocol.io/specification/2025-11-25/basic/transports (read 12 Nov 2026) - modelcontextprotocol.io: connect to local servers: https://modelcontextprotocol.io/docs/develop/connect-local-servers (read 12 Nov 2026) - Cursor: MCP docs: https://cursor.com/docs/context/mcp (read 12 Nov 2026) - Microsoft: VS Code MCP servers: https://code.visualstudio.com/docs/copilot/customization/mcp-servers (read 12 Nov 2026) - Microsoft: VS Code enterprise policies: https://code.visualstudio.com/docs/enterprise/policies (read 12 Nov 2026) - Anthropic: Claude Code MCP: https://code.claude.com/docs/en/mcp (read 12 Nov 2026) - Anthropic: Claude Code managed MCP: https://code.claude.com/docs/en/managed-mcp (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/guides/local-mcp-servers-on-the-endpoint/ --- # Browser agents act with the user's logged-in sessions, so control extension installs by policy and limit where agents may click GUIDE, read 12 Nov 2026 **The short answer.** AI browser extensions and agentic browsers such as Comet, ChatGPT Atlas and Gemini in Chrome read pages and act inside the user's signed-in sessions. Anthropic reported a 1% attack success rate for its browser agent against an adaptive attacker in November 2025. Control extensions through Chrome and Edge policy, and scope agent access by site. ## Two different things get called browser AI, and they need different controls The first is an AI extension: add-on code that runs inside Chrome or Edge and reads or changes pages. Writing assistants, meeting summarizers, page-chat tools and coding helpers are typical. An extension is installed from a store, runs under the permissions in its manifest and can be managed by enterprise policy. The second is an agentic browser, where the AI is built into the browser itself and can act on pages for the user: click, fill forms, move between tabs and carry out multi-step tasks. Perplexity's Comet is one example, and its site lists Mac, Windows, iOS and Android as the platforms (Perplexity, Comet page, read 12 Nov 2026, vendor claim). OpenAI's ChatGPT Atlas and Gemini in Chrome are two others. For those two, the vendor pages we tried returned an error to our fetch on 12 Nov 2026, so this guide makes no product-specific claim about their admin controls. Read each vendor's current enterprise documentation before you decide. The split matters because the control points differ. For an extension, the control point is the browser's extension policy. For an agentic browser, the first control point is whether the browser may be installed at all, since it is a separate application. After that you control the sites it can reach and the accounts it can use. ## Extension permissions define what the add-on can see Chrome shows users a permission warning for powerful APIs. Google's developer documentation explains that these warnings describe the capabilities an API grants, and that some warnings are suppressed when a broader one is present: the warning for the tabs permission does not show when the extension also requests all URLs (Chrome for Developers, permission warnings page, read 12 Nov 2026). The practical reading for a security team is that an extension with access to all URLs can read and modify every page the user opens, including internal tools, webmail and admin consoles. An AI extension usually needs that breadth, because it reads whatever page the user asks about. That is the core tension. The feature that makes the extension useful is the same access that makes a malicious or compromised update dangerous. A trusted extension that is sold to a new owner and updated can start sending page content to a new endpoint without the user doing anything. - Review the manifest permissions for each AI extension: host permissions, tabs, cookies, clipboard, scripting, webRequest. - Check where the extension sends data: the vendor's domain, a model provider, or a third party. Read the privacy policy, and compare it with traffic you can see. - Check who publishes it and how often it updates. A change of publisher is an event worth an alert. - Compare store listing claims with observed behavior on a test profile. ## Chrome and Edge policy can allow, block and force extensions Both browsers manage extensions through policy. Microsoft documents the ExtensionSettings policy for Edge, which can control settings such as the update URL and blocked permissions, and which can override other extension policies set elsewhere in Group Policy, including ExtensionAllowedTypes, ExtensionInstallBlocklist, ExtensionInstallForcelist, ExtensionInstallSources and ExtensionInstallAllowlist (Microsoft Learn, Edge extension management reference, read 12 Nov 2026). Google publishes the same policy for Chrome in its Chrome Enterprise help (Google, Configure ExtensionSettings policy, read 12 Nov 2026). The strongest posture is a default deny: block all extensions, then allow the ones you have reviewed by extension ID. That posture stops an employee from installing a new AI helper from the store on a managed browser profile. A weaker posture, a blocklist of named AI extensions, fails quietly as new ones appear weekly. Two details deserve attention. First, the ExtensionSettings policy can restrict the permissions an allowed extension may use, so you can allow a tool and still block, for example, a host permission you do not want it to hold. Second, Google's help page on Incognito notes that if you install extensions for data protection, you need to make sure users cannot open them in Incognito mode and escape those controls, and it points to the IncognitoModeAvailability policy (Google, Chrome Enterprise Help, read 12 Nov 2026). A control that disappears in a private window is not a control. ## Agentic browsers add actions, and actions bring prompt injection into the browser A chat sidebar reads a page and answers. An agent reads a page and then does something. That difference moves the risk. Text on a page, in an email or in a document can contain instructions aimed at the AI, and if the agent follows them, it acts with the user's logged-in sessions. Anthropic's research post on prompt injection in browser use, published 24 Nov 2025, says it directly: browser use creates unique prompt injection risks, and no browser agent is immune. In that post, Anthropic reported a 1% attack success rate for its newest model against an internal adaptive attacker given 100 attempts per environment, and called 1% a significant improvement that still represents meaningful risk (Anthropic, prompt injection defenses post, read 12 Nov 2026). That is a vendor's measure of its own system, and it comes with the vendor's warning that the problem is not solved. The OWASP guidance on excessive agency frames the same issue for any LLM application. It names three root causes: excessive functionality, excessive permissions and excessive autonomy (OWASP GenAI Security Project, LLM06 excessive agency, read 12 Nov 2026). A browser agent that can act on any site, with every session the user holds, with no confirmation step, has all three. ## The sessions an agent inherits are the real exposure A browser agent does not need your password. It runs in a profile where the user is already signed in to email, the SSO portal, the CRM, the cloud console and the code host. Whatever the user can do in those sessions, the agent can attempt. This is why the browser profile is the unit to think about. A managed work profile with SSO sessions and a personal profile with a consumer AI account are different risks. If an agentic browser is installed next to the managed browser and the user signs in to work apps in it, the work sessions now live in an application that your extension policy, your DLP extension and your network controls may not cover. Check three things on any agentic browser you evaluate: whether it can run in a mode with no access to signed-in sessions, whether it asks for confirmation before sensitive actions such as sending mail, buying or changing settings, and whether administrators can restrict the sites it may act on. Take the answers from current vendor documentation and then confirm them on a test machine. ## Data leaves through paste, upload and page reads, and DLP can see some of it Browser-based AI moves data in three ways: the user pastes text into a prompt, the user uploads a file, or the AI reads the page. Microsoft's Purview documentation states that DLP policies can warn or block users from sharing sensitive information with third-party generative AI sites accessed through a browser, for example stopping a user from pasting credit card numbers into ChatGPT or showing a warning the user can override (Microsoft Learn, Purview and generative AI apps page, read 12 Nov 2026). Purview endpoint DLP lists a paste-to-supported-browsers activity that evaluates the content being pasted, regardless of how its source item was classified (Microsoft Learn, endpoint DLP page, read 12 Nov 2026). Those controls cover paste and upload by the user. A page read by an agent is a different path, since the agent collects content from the page it is on, and the user may never paste anything. The ai-dlp-on-the-endpoint guide on this site covers that gap in detail. ## A control set a security team can deploy this quarter Order the work by effort and effect. - Set extension policy to default deny on managed Chrome and Edge profiles, and keep a reviewed allowlist by extension ID. - Restrict the permissions an allowed AI extension may hold through ExtensionSettings, and set IncognitoModeAvailability so private windows do not bypass extension controls. - Inventory what is installed today on a sample of devices before you enforce, so you know which AI extensions people rely on. - Decide whether agentic browsers are allowed at all. If not, block installation with your application control tooling. If yes, define which accounts may be used in them. - Turn on browser paste and upload DLP for the AI sites you sanction, and warn or block for the rest, using Purview or your DLP product's browser component. - Require confirmation for actions that change data or spend money, wherever the agent product offers the setting. - Log extension installs, removals and permission changes, and alert on a new extension with all-URL host access. ## How to test a vendor that says it covers browser AI Install one AI extension, one agentic browser and one web chat site on a test machine. Ask the vendor's product to show each in its inventory. Paste a test string that matches a sensitive pattern, upload a dummy file and check whether the product records the event and whether it can stop it before it completes. Open the same chat in a second browser and in a private window and check again. A product that sees only the managed browser has said what it covers, even if it did not say so in the data sheet. Then ask what the product does about an agent that reads page content without any paste. If the answer is a recommendation to restrict sites, that is a policy control, and it is worth having. If the answer is that the product observes the agent's actions, ask to see one. ## Where this sits in the index - [Agent Surface Map](/agent-surface-map/): Which of six agent surfaces each of 16 tools sees - [Action Control Matrix](/action-control-matrix/): Which of six agent actions each tool can stop - [The ranked index](/ranking/): 16 tools on seven published criteria ## Related guides - [AI DLP on the endpoint](/guides/ai-dlp-on-the-endpoint/) - [Shadow AI detection signals](/guides/shadow-ai-detection-signals/) - [Local MCP servers on the endpoint](/guides/local-mcp-servers-on-the-endpoint/) ## Sources - Perplexity: Comet: https://www.perplexity.ai/comet (read 12 Nov 2026) - Google: Chrome extension permission warnings: https://developer.chrome.com/docs/extensions/develop/concepts/permission-warnings (read 12 Nov 2026) - Microsoft Learn: Edge extension management reference: https://learn.microsoft.com/en-us/deployedge/microsoft-edge-manage-extensions-ref-guide (read 12 Nov 2026) - Google: Chrome Enterprise Help, ExtensionSettings policy: https://support.google.com/chrome/a/answer/9867568 (read 12 Nov 2026) - Google: Chrome Enterprise Help, extensions in Incognito: https://support.google.com/chrome/a/answer/13130396 (read 12 Nov 2026) - Anthropic: prompt injection defenses in browser use: https://www.anthropic.com/research/prompt-injection-defenses (read 12 Nov 2026) - OWASP GenAI: LLM06 excessive agency: https://genai.owasp.org/llmrisk/llm062025-excessive-agency/ (read 12 Nov 2026) - Microsoft Learn: Purview and generative AI apps: https://learn.microsoft.com/en-us/purview/ai-microsoft-purview (read 12 Nov 2026) - Microsoft Learn: endpoint DLP: https://learn.microsoft.com/en-us/purview/endpoint-dlp-learn-about (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/guides/browser-agents-and-ai-extensions/ --- # AI DLP on the endpoint catches pastes and uploads to AI sites well and misses what agents read and send on their own GUIDE, read 12 Nov 2026 **The short answer.** AI DLP on the endpoint applies data loss prevention to pastes, uploads and agent reads involving AI tools. Microsoft Purview endpoint DLP covers Windows 10/11 and the three latest macOS releases, with browser paste and upload rules. Agents that read files and call tools directly sit outside most classic DLP rules. ## AI DLP is classic DLP pointed at a new set of destinations and a new kind of actor Data loss prevention (DLP) finds sensitive content and decides whether an action may proceed: allow, warn, block or audit. On the endpoint, the actions DLP knows about are the ones people take: copy a file to a USB drive, print, upload to a website, paste into a page. AI changes two things. The destinations are new, because every employee now has a chat box one tab away. And the actor is sometimes software, because an agent can read a folder, call a tool and send the result to a model without a person touching the keyboard. This guide separates three flows that vendors often bundle under the label AI DLP: a person pasting text into an AI tool, a person uploading a file to an AI tool, and an agent reading data and sending it on. The first two are covered by mature controls. The third is where coverage thins out, and where a buyer should ask the most questions. Facts about Microsoft Purview come from Microsoft's documentation, read on 12 Nov 2026. Purview is the best-documented example, and it is not the only product that offers these controls. ## Purview endpoint DLP covers paste and upload to AI sites in the browser On the AI side, Microsoft's Purview page says DLP policies can warn or block users from sharing sensitive information with third-party generative AI sites accessed through a browser. Its example is a user prevented from pasting credit card numbers into ChatGPT, or shown a warning with an override. The same page says AI apps can be detected through browser activity and categorized as Generative AI in the Defender for Cloud Apps catalog (Microsoft Learn, Purview and generative AI apps page). For Chrome on Windows 10 and 11, the Microsoft Purview extension for Chrome is the component that lets the organization monitor attempts to access or upload a sensitive item to a cloud service in Chrome and enforce protective actions (Microsoft Learn, Purview extension for Chrome page). So browser-level paste control for Chrome depends on that extension being installed and not removable by the user. - Upload to a restricted cloud service domain, or access from an unallowed browser. If the user is in an unallowed browser, the upload is blocked and the user is redirected to Microsoft Edge. - Paste to supported browsers. It detects when a user tries to paste content to a restricted service domain, and it evaluates the content being pasted, regardless of how the source item was classified. - Copy to clipboard. For a protected file, the policy can audit, block with override or block. - Browser and domain restrictions through sensitive service domain groups, which let you name which AI sites count as restricted. ## Classic DLP rules assume a person at the keyboard Most DLP policies were written around a user action with a clear start and end: a file is dragged, a button is pressed, a form is submitted. They trigger on content that matches a sensitive information type, a trainable classifier or a sensitivity label. Agents break these assumptions in four ways. The first is volume: an agent reading a repository opens thousands of files, and a rule that audits file opens is either silent or unusable. The second is indirection: the agent reads a file, summarizes it in memory and sends the summary to a model provider over an API call, so the artifact that crosses the network is a paraphrase, not the original document. A content match on the original pattern may not fire. The third is channel: many agent calls go from a terminal process or desktop runtime to an API endpoint, never through a browser form. The fourth is attribution: the audit log says a process made a network call, not that an AI agent decided to send Q3 customer data. OWASP's guidance on excessive agency lists the root causes of agent-driven harm as excessive functionality, excessive permissions and excessive autonomy (OWASP GenAI Security Project, LLM06). Each of the three is a data-loss factor when the agent can read broadly and send freely. ## Agent reads and tool calls need controls at the action, not only at the destination For an agent, the control point is the action: read this file, run this command, call this tool with these arguments. A DLP-aware product for agents should evaluate the data in the call and the destination of the call, then allow, ask or block before the action completes. Some of this exists inside the agent tools themselves. Anthropic documents permission rules and sandboxing for Claude Code, and managed settings that an organization can deploy to endpoints through MDM or a managed file (Anthropic, Claude Code settings and server-managed settings pages, read 12 Nov 2026). Those are guardrails on what the agent may touch. They are not content inspection. They answer whether the agent may read a folder, and they do not answer whether the file contains a customer record. Product claims for content inspection of agent traffic should be read as vendor claims. The questions that test them are in the buyer checks below. ## What each flow needs, and where classic coverage ends The table compares the three flows on what a mature control looks like today. | Flow | Typical control | Where it runs | Known gap | | --- | --- | --- | --- | | Person pastes text into an AI site | Paste rule with warn or block, evaluated on the pasted content | Browser extension or endpoint DLP component | Only browsers the product supports; other apps and agents are outside | | Person uploads a file to an AI site | Upload rule on restricted service domains, with an allowed-browser list | Endpoint DLP plus browser component | Desktop AI apps that do not use a managed browser | | Agent reads files and sends content to a model | Permission rules on file scope, network allowlists, sandboxing | The agent tool's own settings, EDR, or a dedicated agent security product | Content inspection of API traffic is uncommon; paraphrased content evades pattern match | | Agent calls an MCP tool with sensitive arguments | Server allowlists, tool-call policy | Client policy or a product that sees tool calls | Few products show the arguments; check in trial | ## Labels and classifiers help, if the agent's path touches them Sensitivity labels travel with files, and a label can drive a block decision on a copy, upload or paste. They help only where the enforcement point sees the label. Microsoft notes that endpoint DLP cannot detect a sensitivity label from another tenant on a document (Microsoft Learn, endpoint DLP page). If agents read files that carry labels from partners or from a second tenant, the label will not protect the document. Unlabeled data is the larger problem. Source code, internal tickets, customer exports and meeting notes often have no label. Pattern-based types, such as card numbers, catch some of it. A rule for source code or contract text needs a classifier or a keyword set your team maintains. Expect to tune. A rule that fires on every paste of a code snippet will be turned off within a week. ## Warn first on paste, block on upload, and measure before you enforce A sensible rollout sequence follows from how people actually use AI tools. - Start in audit mode for 30 days on the AI sites you sanction and the ones you do not. Count pastes and uploads per site and per sensitive type. - Decide the sanctioned set. For sanctioned sites, allow paste of low-sensitivity content and warn with override on medium. For unsanctioned sites, block high-sensitivity content and offer a link to the approved alternative. - Block uploads of labeled or classified files to unsanctioned AI sites, and set an allowed-browser list so uploads cannot be made from a browser your policy does not reach. - Review override reasons weekly in the first month. They show which rules are wrong and which workflows lack an approved path. - Write the agent rules separately: file scope per project, no read of credential stores, a network allowlist, and an approval step for outbound calls to new domains. Enforce these in the agent tool's managed settings where available. ## Questions that test a vendor's AI DLP claim Ask each question in a trial, using test data you control. Which browsers and which operating systems does the paste rule run in, and what happens in an unsupported browser? Microsoft's own table shows paste support varying by platform, with some entries marked preview, so ask for the current matrix. What does the product see from a desktop AI app that is not a browser? Paste a test string into it and look for the event. What does the product see when a command-line agent reads a file and calls a model API? Ask to see the log line, and ask whether the content of the call is inspected or only the destination. Can the product stop the action before it completes, or does it alert after the data has left? Ask for a demonstration of a block on your machine. What can a user do to disable the control? Try it. How are overrides recorded, and who sees them? A vendor's answer is a vendor claim. On this site's scorecard a claim stays capped at 7 of 10 until a second source confirms it, and a feature with no evidence found scores 0. ## Where this sits in the index - [Agent Surface Map](/agent-surface-map/): Which of six agent surfaces each of 16 tools sees - [Action Control Matrix](/action-control-matrix/): Which of six agent actions each tool can stop - [The ranked index](/ranking/): 16 tools on seven published criteria ## Related guides - [Shadow AI detection signals](/guides/shadow-ai-detection-signals/) - [Agent surfaces explained](/guides/agent-surfaces-explained/) - [Browser agents and AI extensions](/guides/browser-agents-and-ai-extensions/) ## Sources - Microsoft Learn: endpoint DLP: https://learn.microsoft.com/en-us/purview/endpoint-dlp-learn-about (read 12 Nov 2026) - Microsoft Learn: Purview and generative AI apps: https://learn.microsoft.com/en-us/purview/ai-microsoft-purview (read 12 Nov 2026) - Microsoft Learn: Purview extension for Chrome: https://learn.microsoft.com/en-us/purview/dlp-chrome-learn-about (read 12 Nov 2026) - OWASP GenAI: LLM06 excessive agency: https://genai.owasp.org/llmrisk/llm062025-excessive-agency/ (read 12 Nov 2026) - Anthropic: Claude Code settings: https://code.claude.com/docs/en/settings (read 12 Nov 2026) - Anthropic: Claude Code server-managed settings: https://code.claude.com/docs/en/server-managed-settings (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/guides/ai-dlp-on-the-endpoint/ --- # Shadow AI detection works by combining five endpoint signals, because each one sees a different slice and misses the rest GUIDE, read 12 Nov 2026 **The short answer.** Shadow AI detection on endpoints draws on five signals: process and binary inventory, browser extension inventory, network and DNS records, agent and MCP config files, and OAuth grants. Each sees one slice. DNS shows a domain, not a prompt. Combine at least three and test with a known tool. ## Detection is a data-fusion problem, because no single signal is complete Finding unsanctioned AI on laptops is a detection task, and like any detection task it starts from signals. A signal is a record that exists whether or not you are looking: a process table, an extension list, a DNS query, a config file, a consent record. The signals below are the ones that exist on or near the endpoint for AI tools. Each is described by what it sees, what it misses and what it costs to collect. This guide covers the signals. It does not explain why shadow AI happens or how to write the policy that defines it. Both belong in other guides. The reader is assumed to have an EDR or MDM already and wants to know which of its data answers the question. All product-specific facts come from vendor documentation read on 12 Nov 2026. No vendor evidence files for AI endpoint products were available when this guide was written, so no vendor's detection claim is repeated here. ## Process and binary inventory finds installed apps and agents that run The most direct signal is what runs. A process inventory records executable names, paths, signing identity, parent process and command line. Desktop AI apps appear as signed applications. Command-line agents appear as processes started from a terminal, often a Node, Python or native binary that the user installed with a package manager. Local MCP servers appear as child processes of an AI client. On macOS the system source for process and file events is Apple's Endpoint Security framework (Apple developer documentation, read 12 Nov 2026), which security sensors consume. On Windows the equivalent comes from event tracing and driver-level sensors. Your EDR already uses these sources. What the signal sees well: a binary exists and runs, who ran it, when, and what it launched. What it misses: an AI tool that runs in a browser tab leaves no new process. A portable binary renamed by the user defeats name matching, so match on signing identity, hash and parent chain as well. And a process record does not say what data the tool touched unless file and network events are linked to it. Cost: low if your EDR keeps process history, with a catch. Retention is often short, and AI tools are installed once and used for months. Query installed software and application directories as well as running processes, so a tool run weekly is not missed. - Match on: signing identity, bundle ID or install path, not file name alone. - Pair with: a list of known AI client apps and agent CLIs that you update monthly. - Watch for: unsigned or user-writable-path binaries with network connections to model provider domains. ## Browser extension inventory finds add-ons that read pages Every installed extension has an ID, a version, a publisher and a set of permissions. Chrome and Edge both report these to administrators when the browsers are managed. Microsoft's Edge reference describes the ExtensionSettings policy, which controls the update URL and blocked permissions per extension (Microsoft Learn, Edge extension management reference). Google documents the same policy for Chrome (Google, Configure ExtensionSettings policy). What the signal sees well: which AI extensions are installed, which profiles hold them, what permissions they request and when they changed. A new extension with access to all URLs is a high-value alert, and Google's permission-warnings documentation explains why: that access covers every site the user opens (Chrome for Developers, permission warnings page). What it misses: browsers and profiles outside management, extensions in a second browser you do not collect from, and AI features built into a browser rather than installed as an add-on. It also does not show whether the extension was used, only that it exists. Cost: low for managed browsers through cloud management or policy reporting. Higher for unmanaged profiles, where you need a collector on the machine. - Match on: extension ID against a maintained list, plus keyword match on name and description for new entries. - Alert on: a host permission change, an update-URL change or a publisher change for an allowed extension. - Cross-check: the inventory against your allowlist to find installs that policy should have blocked. ## Network and DNS records find use of AI services, with weak attribution DNS queries and network connections to model provider domains show that something on a machine talked to an AI service. A resolver or proxy log is the cheapest place to see this at scale, because it covers every device on the network and needs no endpoint software. Microsoft's approach illustrates the idea at the browser level: Purview documents that AI apps can be detected through browser activity and categorized as Generative AI in the Defender for Cloud Apps catalog (Microsoft Learn, Purview and generative AI apps page). A categorized domain list is the building block, whoever supplies it. What the signal sees well: that a device resolved or connected to an AI domain, how often, and at what volume of bytes. It also finds tools you did not know to look for, because a new domain shows up in the categorization even when no process match exists. What it misses is a long list. It does not show prompts, files or tool calls. It does not name the process, unless the log comes from an endpoint sensor that records the process with each connection. Encrypted DNS and VPNs hide it from network-side logs. Traffic off the corporate network, on a home connection, never reaches your resolver. And many AI features sit behind domains shared with general services, so a domain match gives a false positive or a false negative. Cost: low at the resolver or proxy, medium on the endpoint. For attribution, collect connection records with process name and user on the device itself. - Use for: breadth and discovery of new services. - Do not use for: proof of data exposure or of a specific user action. - Improve with: process-linked connection records and a domain list that you review monthly. ## Agent and MCP config files find what an AI tool is set up to do Config files describe intent. They are on disk before the tool runs and stay after it stops. For MCP, each client keeps its server list in a defined place: Claude Desktop in claude_desktop_config.json under the user's application support or APPDATA folder (modelcontextprotocol.io, connect local servers page), Cursor in .cursor/mcp.json and ~/.cursor/mcp.json (Cursor docs), and Claude Code in .mcp.json and ~/.claude.json (Anthropic, Claude Code MCP page). The sibling guide on local MCP servers lists the paths in a table. What the signal sees well: which servers a user configured, the command each runs, the arguments, and which environment variable names are set. It finds tools that are installed and dormant. It also finds project-level definitions that arrived with a cloned repository. What it misses: it shows configuration, not use. A server may be defined and never launched. A tool whose config lives in a location you do not know to scan is invisible. Cloud-hosted agents have configs on a server, not on the laptop. And a config file is easy to edit, so file-integrity monitoring matters more than a one-time scan. Cost: low to medium. You need a script or sensor that can read files in every user profile, extract fields and exclude secret values. - Collect: path, hash, modified time, server names, command and args, env variable names (not values). - Alert on: a new server, a changed command, or a server defined with a package runner such as npx. - Update: the path list as clients ship new versions. ## OAuth grants find AI apps connected to corporate data, whatever device they run on When a user signs in to an AI app with a work account, or authorizes an app to read mail or files, the identity provider records the consent. Microsoft documents how to review the permissions granted to applications in a Microsoft Entra tenant, covering applications added through user or admin consent, and the roles needed to do it (Microsoft Learn, manage application permissions page). What the signal sees well: which AI apps hold access to corporate data, which scopes they hold, which users consented, and when. It works for apps on unmanaged devices and phones, because the grant lives in the tenant. That makes it the best signal for the cases where nothing runs on a managed laptop. What it misses: tools that never touch your identity provider. A personal account on a consumer AI service, an API key pasted into a command-line agent, and a local model need no OAuth grant. It also shows the app, not the activity: a grant says the app can read mail, not that it did. Cost: low. It is an administrative query, and it can run daily. - Review: grants with write or full-mailbox scopes first. - Alert on: a new AI app consented by multiple users in one week. - Revoke: grants for apps outside your approved list, after notifying users. ## The five signals side by side The table compares what each signal can establish. 'Presence' means the tool exists on a device or in a tenant. 'Use' means evidence it ran. 'Data' means evidence of what content it handled. | Signal | Presence | Use | Data | Main blind spot | | --- | --- | --- | --- | --- | | Process and binary inventory | Yes | Yes | Only if linked to file and network events | Browser-only AI tools | | Extension inventory | Yes | Not shown | No | Unmanaged browsers and profiles | | Network and DNS | Indirect | Yes, by domain | No | Prompts, files, and traffic off the network | | Config files | Yes, including dormant tools | Not shown | No | Unknown paths and cloud-hosted agents | | OAuth grants | Yes, for tenant-connected apps | Partial | Scopes only | Personal accounts and API keys | ## Combine signals into one finding per tool and test the pipeline with a known app Raw signals produce noise. The useful unit is one finding per tool per user: this person has this AI tool, found by these signals, last seen on this date. A finding supported by two separate signals, such as a process and a DNS record, is more reliable than either alone. A finding supported by a config file alone is a possibility, not a confirmed use. Test the pipeline before you trust it. Install a tool you know, such as an AI extension on a test profile, a command-line agent in a terminal and a local MCP server in a config file. Check which signals fired, how long each took to appear and whether they merged into one finding. Remove the tool and check that the finding ages out. If a tool you installed on purpose does not appear, you have measured your blind spot cheaply. When you evaluate a product that claims shadow AI discovery, ask which of the five signals it collects, from which component, and how it joins them. A product that lists one signal has told you its blind spot. The scorecard on this site treats a claim without a confirming source as a vendor claim, capped at 7 of 10. ## Where this sits in the index - [Agent Surface Map](/agent-surface-map/): Which of six agent surfaces each of 16 tools sees - [Action Control Matrix](/action-control-matrix/): Which of six agent actions each tool can stop - [The ranked index](/ranking/): 16 tools on seven published criteria ## Related guides - [Agent surfaces explained](/guides/agent-surfaces-explained/) - [Securing Claude Code on developer laptops](/guides/claude-code-security/) - [AI DLP on the endpoint](/guides/ai-dlp-on-the-endpoint/) ## Sources - Apple: Endpoint Security framework: https://developer.apple.com/documentation/endpointsecurity (read 12 Nov 2026) - Microsoft Learn: Edge extension management reference: https://learn.microsoft.com/en-us/deployedge/microsoft-edge-manage-extensions-ref-guide (read 12 Nov 2026) - Google: Chrome Enterprise Help, ExtensionSettings policy: https://support.google.com/chrome/a/answer/9867568 (read 12 Nov 2026) - Google: Chrome extension permission warnings: https://developer.chrome.com/docs/extensions/develop/concepts/permission-warnings (read 12 Nov 2026) - Microsoft Learn: Purview and generative AI apps: https://learn.microsoft.com/en-us/purview/ai-microsoft-purview (read 12 Nov 2026) - modelcontextprotocol.io: connect to local servers: https://modelcontextprotocol.io/docs/develop/connect-local-servers (read 12 Nov 2026) - Cursor: MCP docs: https://cursor.com/docs/context/mcp (read 12 Nov 2026) - Anthropic: Claude Code MCP: https://code.claude.com/docs/en/mcp (read 12 Nov 2026) - Microsoft Learn: manage application permissions: https://learn.microsoft.com/en-us/entra/identity/enterprise-apps/manage-application-permissions (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/guides/shadow-ai-detection-signals/ --- # Six agent surfaces cover AI on an employee laptop, and each one runs different code with different privileges GUIDE, read 12 Nov 2026 **The short answer.** Six agent surfaces make up AI on a laptop: desktop AI apps, IDE assistants, command-line agents, MCP servers, browser extensions and SaaS copilots. They differ in where they run and what they can touch. Command-line agents and local MCP servers run as the user, with shell and file access. SaaS copilots run in the vendor's cloud. ## How does endpoint security work for AI agents? An AI agent is software that takes actions on a person's behalf, usually by calling tools in a loop. Where that software runs decides which actions it can take and which controls can reach it. A program in a browser tab is bound by the browser's rules. A program started from a terminal is bound by the user's operating system permissions. A service in a vendor's cloud is bound by the permissions you granted it through OAuth. ## A surface is a place where an AI system acts, and the place sets the privileges This site groups agent software into six surfaces. The grouping is by place of execution, not by vendor, because the endpoint control that applies depends on the place. Examples named below are illustrations from vendor documentation read on 12 Nov 2026, and the list is not a market map. The six actions that matter for security are file read, shell command, network call, credential access, data paste and package install. The last section crosses the surfaces with the actions. Reading that table first is a reasonable way to use this page. ## Desktop AI apps run as the user, but usually inside their own application boundary A desktop AI app is a standalone application with a chat interface, installed like any other program. Claude Desktop is one. It reads a config file that can define local MCP servers, in claude_desktop_config.json under the user's application support (macOS), APPDATA (Windows) or .config (Linux) folder (modelcontextprotocol.io, connect local servers page). What runs: the app's own process, plus any child processes it starts for tools. What privileges: those of the signed-in user, limited by the operating system's per-app permission prompts, such as access to the Documents folder on macOS. What matters most: the app reads files the user attaches, and with MCP servers configured it can read and write files, query systems and run commands through them. Control points: application allowlisting, MDM-managed settings where the app supports them, and the config file that adds tools. The app's own network traffic goes to its provider, and a network proxy can see the domain. ## IDE assistants run inside the editor and inherit the project's reach IDE assistants live in an editor or an editor fork. GitHub Copilot in VS Code and Cursor are two examples. They read open files and the project tree, propose edits and, in agent mode, run terminal commands and call MCP tools. What runs: the editor process and its extension host, plus terminals and tool servers it launches. What privileges: the user's. An editor terminal is a normal shell. What matters most: file read across the workspace, shell command execution, package install through the terminal, and credential access through files such as environment files in the project. Control points are richer than for desktop apps. Microsoft publishes VS Code enterprise policies that control access to MCP servers and an allowlist for them, along with policies for agent hooks and plugins (Microsoft, VS Code enterprise policies page). Cursor documents its own MCP config files and fields (Cursor docs, MCP). Because these are editors, they usually run on developer machines that hold source code, signing keys and cloud credentials, which raises the stakes. ## Command-line agents run in a terminal with the user's shell, files and credentials A command-line agent is started from a shell and works on the current directory. Claude Code is the documented example here. It has user, project and organization scopes for settings and MCP servers, and an organization can deploy endpoint-managed settings through native OS policies (macOS managed preferences, Windows registry) or a managed settings file (Anthropic, Claude Code server-managed settings page). What runs: a long-lived process in a terminal, which starts child processes for each shell command. What privileges: everything the user's shell can do, bounded by whatever permission rules and sandboxing the agent applies. What matters most: shell command, file read and write, network call, package install and credential access, since the shell can reach the user's SSH keys, cloud credentials and tokens. This is the surface with the thinnest default guardrails outside the tool's own settings, and the one where a process record is least informative, since a shell command run by the agent looks like one typed by the developer. Control points are the agent's managed settings, sandboxing where available, and an endpoint sensor that links commands to the agent session. ## MCP servers are separate processes or remote services that give agents tools The Model Context Protocol defines how an AI application calls tools and data sources. A server can run locally as a subprocess that the client launches over stdio, or remotely over Streamable HTTP (Model Context Protocol specification, transports page, read 12 Nov 2026). A local server can be a file system tool, a database client, a browser controller or a wrapper around a company API. What runs: a process on the laptop, started by the client from a command in a config file. What privileges: the MCP security best practices page says local servers run with the same privileges as the client and may have direct access to the user's system, and it asks clients to show the exact startup command, require consent, and sandbox the server with restricted file system and network access (modelcontextprotocol.io, security best practices page). What matters most: whatever the server exposes. A file server gives file read and write. A shell server gives shell commands. A database server gives network calls with stored credentials. Because one client config can define many servers, the surface multiplies the actions of every agent that uses it. The local MCP servers guide covers discovery and control in detail. ## Browser extensions and AI browsers run inside the browser with the user's sessions Browser extensions run in Chrome or Edge under the permissions in their manifest. Google explains that permission warnings describe API capabilities, and that a broad host permission lets an extension act across all sites (Chrome for Developers, permission warnings page). AI browsers such as Perplexity's Comet embed the AI in the browser itself (Perplexity, Comet page, vendor claim). What runs: extension code in the browser, or an agent inside an AI browser. What privileges: the page content and the user's signed-in sessions, but not the local shell or arbitrary files, with the exception of file pickers the user approves. What matters most: data paste and upload, network calls to the AI provider, and actions on sites where the user is signed in. Control points are the extension policy in Chrome and Edge, which can allow, block, force and restrict extensions (Microsoft Learn, Edge extension management reference; Google, Configure ExtensionSettings policy), and browser-level DLP. Anthropic's note that no browser agent is immune to prompt injection applies to this surface in particular (Anthropic, prompt injection defenses post). ## SaaS copilots run in the vendor's cloud, with permissions you granted A SaaS copilot is an AI feature inside a hosted product: Microsoft 365 Copilot is one, and Microsoft documents the Purview protections that apply to it and to other generative AI apps (Microsoft Learn, Purview and generative AI apps page). It runs in the vendor's cloud and acts on the data the signed-in user can already access in that product. What runs: nothing on the laptop other than the browser or app used to reach it. What privileges: the user's permissions inside the SaaS product, plus any connectors an admin turned on. What matters most: data the copilot can retrieve and summarize, which is everything the user can open. The risk is oversharing: a user with access to too many files lets the copilot surface them. Control points are tenant settings, sensitivity labels, access reviews and the OAuth grants for connected apps (Microsoft Learn, manage application permissions page). An endpoint product sees this surface only through the browser or through an API connector. ## Which actions each surface can take, side by side The table marks what each surface can typically do. 'Yes' means the capability is common in the category. Individual products differ, and tools often gate these behind permission prompts that a user or policy can set. | Surface | File read | Shell command | Network call | Credential access | Data paste | Package install | | --- | --- | --- | --- | --- | --- | --- | | Desktop AI apps | User-attached files; more through MCP | Through MCP tools | Yes, to the provider and tool servers | Through MCP config and env values | Yes | Through MCP tools | | IDE assistants | Yes, across the workspace | Yes, in agent mode | Yes | Project files and environment | Yes | Yes, via the terminal | | Command-line agents | Yes | Yes | Yes | Yes, via the user's shell and files | Not a typical path | Yes | | MCP servers | Whatever the server exposes | If the server exposes it | Yes | Env values in config | Not a typical path | Package runners at launch | | Browser extensions | Page content only | No | Yes | Cookies and sessions, by permission | Yes | No | | SaaS copilots | Cloud data the user can open | No | Within the vendor's cloud | Via tenant connectors | Yes, in the prompt | No | ## Use the surfaces to ask a vendor where its coverage stops Take the six surfaces and the six actions and put a mark next to each pair for which a vendor claims coverage. Then ask the vendor for the component that sees it, and the way it can stop it. Most products cover two or three surfaces well. A product that claims all six should show you each one running on a test device. Order your own priorities by exposure. For a company with many developers, command-line agents, IDE assistants and MCP servers carry the highest-impact actions: shell commands, credential access and package installs. For a company where most staff are knowledge workers, browser extensions, desktop AI apps and SaaS copilots hold more of the data-paste and oversharing risk. OWASP's guidance on excessive agency is a useful frame for the review. It names excessive functionality, excessive permissions and excessive autonomy as root causes (OWASP GenAI Security Project, LLM06). Each surface above is a place where one of the three can creep in: a tool server with more functions than needed, a shell with broad permissions, a browser agent that acts without confirmation. ## Where this sits in the index - [Agent Surface Map](/agent-surface-map/): Which of six agent surfaces each of 16 tools sees - [Action Control Matrix](/action-control-matrix/): Which of six agent actions each tool can stop - [The ranked index](/ranking/): 16 tools on seven published criteria ## Related guides - [Securing Claude Code on developer laptops](/guides/claude-code-security/) - [Securing Cursor on developer machines](/guides/cursor-security/) - [Shadow AI detection signals](/guides/shadow-ai-detection-signals/) ## Sources - modelcontextprotocol.io: connect to local servers: https://modelcontextprotocol.io/docs/develop/connect-local-servers (read 12 Nov 2026) - modelcontextprotocol.io: transports specification 2025-11-25: https://modelcontextprotocol.io/specification/2025-11-25/basic/transports (read 12 Nov 2026) - modelcontextprotocol.io: security best practices: https://modelcontextprotocol.io/docs/tutorials/security/security_best_practices (read 12 Nov 2026) - Microsoft: VS Code enterprise policies: https://code.visualstudio.com/docs/enterprise/policies (read 12 Nov 2026) - Cursor: MCP docs: https://cursor.com/docs/context/mcp (read 12 Nov 2026) - Anthropic: Claude Code server-managed settings: https://code.claude.com/docs/en/server-managed-settings (read 12 Nov 2026) - Google: Chrome extension permission warnings: https://developer.chrome.com/docs/extensions/develop/concepts/permission-warnings (read 12 Nov 2026) - Perplexity: Comet: https://www.perplexity.ai/comet (read 12 Nov 2026) - Microsoft Learn: Edge extension management reference: https://learn.microsoft.com/en-us/deployedge/microsoft-edge-manage-extensions-ref-guide (read 12 Nov 2026) - Google: Chrome Enterprise Help, ExtensionSettings policy: https://support.google.com/chrome/a/answer/9867568 (read 12 Nov 2026) - Anthropic: prompt injection defenses in browser use: https://www.anthropic.com/research/prompt-injection-defenses (read 12 Nov 2026) - Microsoft Learn: Purview and generative AI apps: https://learn.microsoft.com/en-us/purview/ai-microsoft-purview (read 12 Nov 2026) - Microsoft Learn: manage application permissions: https://learn.microsoft.com/en-us/entra/identity/enterprise-apps/manage-application-permissions (read 12 Nov 2026) - OWASP GenAI: LLM06 excessive agency: https://genai.owasp.org/llmrisk/llm062025-excessive-agency/ (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/guides/agent-surfaces-explained/ --- # 12 guides to securing coding agents and the endpoints they run on GUIDES, read 12 Nov 2026 **The short answer.** The guides cover securing coding agents on developer laptops and getting control of AI agents on endpoints: a first inventory, what agentless means, local MCP servers, browser agents, writing an agent policy and where EDR stops. Each guide cites its sources and the date they were read. - [Securing Claude Code on developer laptops](/guides/claude-code-security/): Claude Code ships a permission system, an optional OS sandbox (Seatbelt on macOS, bubblewrap on Linux), hooks, MCP allowlists and OpenTelemetry export. - [Securing Cursor on developer machines](/guides/cursor-security/): Cursor's agent reads files and edits workspace files without approval, asks before shell commands and MCP calls, and offers three Run Modes plus a sandbox. - [Securing GitHub Copilot agent mode and the cloud agent](/guides/github-copilot-agent-mode-security/): Copilot agent mode in VS Code approves terminal commands, MCP tools and URLs, but network access defaults to on (chat.agent.sandbox.network.allowNetwork is true). - [Securing OpenAI Codex CLI on developer laptops](/guides/codex-cli-security/): OpenAI's Codex CLI runs commands in an OS-enforced sandbox with network off and writes limited to the workspace by default, plus an approval policy for leaving it. - [Securing Google Gemini CLI on developer laptops](/guides/gemini-cli-security/): Google's Gemini CLI documents sandboxing through Seatbelt, Docker, Podman, gVisor and a Windows native option, but tools.sandbox defaults to undefined and Trusted Folders is disabled by default. - [What coding agents can touch on a developer laptop, and which controls ship with each](/guides/coding-agents-on-developer-laptops/): Claude Code, Cursor, GitHub Copilot agent mode, Codex CLI and Gemini CLI all run shell commands as the signed-in developer. - [What agentless means on an endpoint](/guides/what-agentless-means-on-an-endpoint/): Agentless AI endpoint security can mean an MDM-pushed profile, a browser extension, an OS API sensor, existing EDR telemetry or a SaaS API connector. - [Local MCP servers on the endpoint](/guides/local-mcp-servers-on-the-endpoint/): A local MCP server is a process an AI client starts on a laptop, usually over stdio, with the user's own file and network rights. - [Browser agents and AI extensions](/guides/browser-agents-and-ai-extensions/): AI browser extensions and agentic browsers such as Comet, ChatGPT Atlas and Gemini in Chrome read pages and act inside the user's signed-in sessions. - [AI DLP on the endpoint](/guides/ai-dlp-on-the-endpoint/): AI DLP on the endpoint applies data loss prevention to pastes, uploads and agent reads involving AI tools. - [Shadow AI detection signals](/guides/shadow-ai-detection-signals/): Shadow AI detection on endpoints draws on five signals: process and binary inventory, browser extension inventory, network and DNS records, agent and MCP config files, and OAuth grants. - [Agent surfaces explained](/guides/agent-surfaces-explained/): Six agent surfaces make up AI on a laptop: desktop AI apps, IDE assistants, command-line agents, MCP servers, browser extensions and SaaS copilots. Short answers to 30 common questions are on the [FAQ page](/faq/). ## Sources - Method: https://aiendpointsecurityranked.com/method/ (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/guides/ --- # AI endpoint security, defined with a sourced example GLOSSARY, read 12 Nov 2026 **The short answer.** AI endpoint security is the practice of finding, governing and stopping the AI tools that run on employee devices. Lasso's guide on securing AI on endpoints says Claude Code or Cursor looks like an approved application to an EDR agent, which cannot see the connected MCP servers or the data in the agent's context. ## Definition AI endpoint security is the practice of finding, governing and stopping the AI tools that run on employee devices. That covers coding agents, desktop AI apps, browser extensions and the MCP servers and skills they connect to. It sits beside EDR, which watches processes and files but not the intent behind an agent's action. ## Example from a real source **Example.** Lasso's guide on securing AI on endpoints says Claude Code or Cursor looks like an approved application to an EDR agent, which cannot see the connected MCP servers or the data in the agent's context. ## Related terms - [Agentic endpoint](/glossary/agentic-endpoint/): An agentic endpoint is a laptop, workstation or developer machine where autonomous AI agents act with the user's access. - [Coding agent](/glossary/coding-agent/): A coding agent is an AI tool that reads a repository, edits files, runs commands and calls tools to finish a development task with limited step-by-step direction. - [YOLO / auto-approve mode](/glossary/yolo-auto-approve-mode/): Auto-approve mode, often called YOLO mode, lets an agent run actions without asking each time. ## Where to read more - [The Agent Surface Map](/agent-surface-map/) - [The Action Control Matrix](/action-control-matrix/) - [All glossary terms](/glossary/) ## Sources - www.lasso.security/blog/securing-ai-on-endpoints-how-to-govern-ai-tools-coding-agents-and-mcp-connections: https://www.lasso.security/blog/securing-ai-on-endpoints-how-to-govern-ai-tools-coding-agents-and-mcp-connections (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/glossary/ai-endpoint-security/ --- # Agentic endpoint, defined with a sourced example GLOSSARY, read 12 Nov 2026 **The short answer.** An agentic endpoint is a laptop, workstation or developer machine where autonomous AI agents act with the user's access. The device is no longer only a place where a person works. Palo Alto Networks' Cortex Agentic Endpoint Security page says that as autonomous AI agents become standard workspace tools, the traditional endpoint has evolved into a new security boundary. ## Definition An agentic endpoint is a laptop, workstation or developer machine where autonomous AI agents act with the user's access. The device is no longer only a place where a person works. It becomes a place where software acts on the person's behalf, with file, shell, network and credential reach. ## Example from a real source **Example.** Palo Alto Networks' Cortex Agentic Endpoint Security page says that as autonomous AI agents become standard workspace tools, the traditional endpoint has evolved into a new security boundary. ## Related terms - [Coding agent](/glossary/coding-agent/): A coding agent is an AI tool that reads a repository, edits files, runs commands and calls tools to finish a development task with limited step-by-step direction. - [Command-line agent](/glossary/command-line-agent/): A command-line agent is a coding agent that runs in a terminal session instead of an editor window. - [AI endpoint security](/glossary/ai-endpoint-security/): AI endpoint security is the practice of finding, governing and stopping the AI tools that run on employee devices. ## Where to read more - [The Agent Surface Map](/agent-surface-map/) - [The Action Control Matrix](/action-control-matrix/) - [All glossary terms](/glossary/) ## Sources - www.paloaltonetworks.com/cortex/agentic-endpoint-security: https://www.paloaltonetworks.com/cortex/agentic-endpoint-security (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/glossary/agentic-endpoint/ --- # Coding agent, defined with a sourced example GLOSSARY, read 12 Nov 2026 **The short answer.** A coding agent is an AI tool that reads a repository, edits files, runs commands and calls tools to finish a development task with limited step-by-step direction. Lasso lists coding agents such as Claude Code and Cursor as a main class of AI running directly on employee devices. ## Definition A coding agent is an AI tool that reads a repository, edits files, runs commands and calls tools to finish a development task with limited step-by-step direction. It runs with the developer's permissions, which is why its approvals, sandbox and network access are security settings and not only usability settings. ## Example from a real source **Example.** Lasso lists coding agents such as Claude Code and Cursor as a main class of AI running directly on employee devices. ## Related terms - [Command-line agent](/glossary/command-line-agent/): A command-line agent is a coding agent that runs in a terminal session instead of an editor window. - [MCP server (local stdio)](/glossary/mcp-server-local-stdio/): A local stdio MCP server is a program the MCP client launches as a subprocess on the same machine, talking over standard input and output. - [Agentic endpoint](/glossary/agentic-endpoint/): An agentic endpoint is a laptop, workstation or developer machine where autonomous AI agents act with the user's access. ## Where to read more - [The Agent Surface Map](/agent-surface-map/) - [The Action Control Matrix](/action-control-matrix/) - [All glossary terms](/glossary/) ## Sources - www.lasso.security/blog/securing-ai-on-endpoints-how-to-govern-ai-tools-coding-agents-and-mcp-connections: https://www.lasso.security/blog/securing-ai-on-endpoints-how-to-govern-ai-tools-coding-agents-and-mcp-connections (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/glossary/coding-agent/ --- # Command-line agent, defined with a sourced example GLOSSARY, read 12 Nov 2026 **The short answer.** A command-line agent is a coding agent that runs in a terminal session instead of an editor window. OpenAI's Codex documentation says that locally Codex uses an OS-enforced sandbox, typically limited to the current workspace, plus an approval policy, and that network access is off by default. ## Definition A command-line agent is a coding agent that runs in a terminal session instead of an editor window. It starts shell commands and edits files on the local machine, so its sandbox and approval policy define how much it can do unattended. ## Example from a real source **Example.** OpenAI's Codex documentation says that locally Codex uses an OS-enforced sandbox, typically limited to the current workspace, plus an approval policy, and that network access is off by default. ## Related terms - [MCP server (local stdio)](/glossary/mcp-server-local-stdio/): A local stdio MCP server is a program the MCP client launches as a subprocess on the same machine, talking over standard input and output. - [MCP client](/glossary/mcp-client/): An MCP client is the connector inside an AI application that holds a dedicated connection to one MCP server. - [Coding agent](/glossary/coding-agent/): A coding agent is an AI tool that reads a repository, edits files, runs commands and calls tools to finish a development task with limited step-by-step direction. ## Where to read more - [The Agent Surface Map](/agent-surface-map/) - [The Action Control Matrix](/action-control-matrix/) - [All glossary terms](/glossary/) ## Sources - developers.openai.com/codex/agent-approvals-security: https://developers.openai.com/codex/agent-approvals-security (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/glossary/command-line-agent/ --- # MCP server (local stdio), defined with a sourced example GLOSSARY, read 12 Nov 2026 **The short answer.** A local stdio MCP server is a program the MCP client launches as a subprocess on the same machine, talking over standard input and output. The glossary defines 36 terms used on this site, each with a sourced example read on 12 Nov 2026. ## Definition A local stdio MCP server is a program the MCP client launches as a subprocess on the same machine, talking over standard input and output. It runs with the user's permissions, so it can see the same files and environment variables as the developer. Treat installing one like installing a binary. ## Example from a real source **Example.** The MCP specification says that in the stdio transport the client launches the server as a subprocess, and its security guidance says local servers are binaries downloaded and executed on the same machine as the client. ## Related terms - [MCP client](/glossary/mcp-client/): An MCP client is the connector inside an AI application that holds a dedicated connection to one MCP server. - [Tool call](/glossary/tool-call/): A tool call is a request from the model, run through the client, to execute a named function on a server or in the agent itself, with arguments. - [Command-line agent](/glossary/command-line-agent/): A command-line agent is a coding agent that runs in a terminal session instead of an editor window. ## Where to read more - [The Agent Surface Map](/agent-surface-map/) - [The Action Control Matrix](/action-control-matrix/) - [All glossary terms](/glossary/) ## Sources - modelcontextprotocol.io/specification/2025-06-18/basic/transports: https://modelcontextprotocol.io/specification/2025-06-18/basic/transports (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/glossary/mcp-server-local-stdio/ --- # MCP client, defined with a sourced example GLOSSARY, read 12 Nov 2026 **The short answer.** An MCP client is the connector inside an AI application that holds a dedicated connection to one MCP server. The application that coordinates clients is called the host. The MCP architecture overview says a host such as Claude Code or Claude Desktop creates one MCP client for each MCP server it connects to. ## Definition An MCP client is the connector inside an AI application that holds a dedicated connection to one MCP server. The application that coordinates clients is called the host. Approval prompts, server allowlists and transport handling are implemented at this layer. ## Example from a real source **Example.** The MCP architecture overview says a host such as Claude Code or Claude Desktop creates one MCP client for each MCP server it connects to. ## Related terms - [Tool call](/glossary/tool-call/): A tool call is a request from the model, run through the client, to execute a named function on a server or in the agent itself, with arguments. - [Agent skill](/glossary/agent-skill/): An agent skill is a packaged set of instructions, and sometimes scripts, that an agent loads when a task matches. - [MCP server (local stdio)](/glossary/mcp-server-local-stdio/): A local stdio MCP server is a program the MCP client launches as a subprocess on the same machine, talking over standard input and output. ## Where to read more - [The Agent Surface Map](/agent-surface-map/) - [The Action Control Matrix](/action-control-matrix/) - [All glossary terms](/glossary/) ## Sources - modelcontextprotocol.io/docs/learn/architecture: https://modelcontextprotocol.io/docs/learn/architecture (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/glossary/mcp-client/ --- # Tool call, defined with a sourced example GLOSSARY, read 12 Nov 2026 **The short answer.** A tool call is a request from the model, run through the client, to execute a named function on a server or in the agent itself, with arguments. The MCP tools specification shows clients invoking a tool with a tools/call request that carries the tool name and arguments. ## Definition A tool call is a request from the model, run through the client, to execute a named function on a server or in the agent itself, with arguments. It is the unit where an agent's words turn into actions, so it is where approvals, logging and blocking attach. ## Example from a real source **Example.** The MCP tools specification shows clients invoking a tool with a tools/call request that carries the tool name and arguments. ## Related terms - [Agent skill](/glossary/agent-skill/): An agent skill is a packaged set of instructions, and sometimes scripts, that an agent loads when a task matches. - [Plugin](/glossary/plugin/): In a coding agent, a plugin is a bundle that can contain skills, sub-agents, hooks and MCP servers, installed as one unit, usually from a marketplace. - [MCP client](/glossary/mcp-client/): An MCP client is the connector inside an AI application that holds a dedicated connection to one MCP server. ## Where to read more - [The Agent Surface Map](/agent-surface-map/) - [The Action Control Matrix](/action-control-matrix/) - [All glossary terms](/glossary/) ## Sources - modelcontextprotocol.io/specification/2025-06-18/server/tools: https://modelcontextprotocol.io/specification/2025-06-18/server/tools (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/glossary/tool-call/ --- # Agent skill, defined with a sourced example GLOSSARY, read 12 Nov 2026 **The short answer.** An agent skill is a packaged set of instructions, and sometimes scripts, that an agent loads when a task matches. Anthropic's Claude Code documentation says you create a SKILL.md file with instructions and Claude adds it to its toolkit, using it when relevant or when invoked directly. ## Definition An agent skill is a packaged set of instructions, and sometimes scripts, that an agent loads when a task matches. Because the agent follows the instructions as if they were its own, a skill from an untrusted source is a way to steer it. ## Example from a real source **Example.** Anthropic's Claude Code documentation says you create a SKILL.md file with instructions and Claude adds it to its toolkit, using it when relevant or when invoked directly. ## Related terms - [Plugin](/glossary/plugin/): In a coding agent, a plugin is a bundle that can contain skills, sub-agents, hooks and MCP servers, installed as one unit, usually from a marketplace. - [Browser agent](/glossary/browser-agent/): A browser agent is an AI agent that navigates web pages, reads content, fills forms and clicks on behalf of a user. - [Tool call](/glossary/tool-call/): A tool call is a request from the model, run through the client, to execute a named function on a server or in the agent itself, with arguments. ## Where to read more - [The Agent Surface Map](/agent-surface-map/) - [The Action Control Matrix](/action-control-matrix/) - [All glossary terms](/glossary/) ## Sources - code.claude.com/docs/en/skills: https://code.claude.com/docs/en/skills (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/glossary/agent-skill/ --- # Plugin, defined with a sourced example GLOSSARY, read 12 Nov 2026 **The short answer.** In a coding agent, a plugin is a bundle that can contain skills, sub-agents, hooks and MCP servers, installed as one unit, usually from a marketplace. Anthropic defines a Claude Code plugin as a directory of skills, agents, hooks, MCP servers or other components that Claude Code installs and loads as one unit. ## Definition In a coding agent, a plugin is a bundle that can contain skills, sub-agents, hooks and MCP servers, installed as one unit, usually from a marketplace. A single install can therefore add several kinds of executable behavior. ## Example from a real source **Example.** Anthropic defines a Claude Code plugin as a directory of skills, agents, hooks, MCP servers or other components that Claude Code installs and loads as one unit. ## Related terms - [Browser agent](/glossary/browser-agent/): A browser agent is an AI agent that navigates web pages, reads content, fills forms and clicks on behalf of a user. - [Shadow AI](/glossary/shadow-ai/): Shadow AI is the use of AI tools that security and IT have not approved or cannot see. - [Agent skill](/glossary/agent-skill/): An agent skill is a packaged set of instructions, and sometimes scripts, that an agent loads when a task matches. ## Where to read more - [The Agent Surface Map](/agent-surface-map/) - [The Action Control Matrix](/action-control-matrix/) - [All glossary terms](/glossary/) ## Sources - code.claude.com/docs/en/plugins: https://code.claude.com/docs/en/plugins (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/glossary/plugin/ --- # Browser agent, defined with a sourced example GLOSSARY, read 12 Nov 2026 **The short answer.** A browser agent is an AI agent that navigates web pages, reads content, fills forms and clicks on behalf of a user. Island's post on its enterprise browser for agents says agents need a browser to work in web and SaaS applications, and that consumer browsers with AI provider extensions have no policy layer. ## Definition A browser agent is an AI agent that navigates web pages, reads content, fills forms and clicks on behalf of a user. It acts through the user's logged-in sessions, so browser policy has to tell agent activity from human activity. ## Example from a real source **Example.** Island's post on its enterprise browser for agents says agents need a browser to work in web and SaaS applications, and that consumer browsers with AI provider extensions have no policy layer. ## Related terms - [Shadow AI](/glossary/shadow-ai/): Shadow AI is the use of AI tools that security and IT have not approved or cannot see. - [AI DLP](/glossary/ai-dlp/): AI DLP is data loss prevention aimed at prompts, uploads and agent context sent to AI tools. - [Plugin](/glossary/plugin/): In a coding agent, a plugin is a bundle that can contain skills, sub-agents, hooks and MCP servers, installed as one unit, usually from a marketplace. ## Where to read more - [The Agent Surface Map](/agent-surface-map/) - [The Action Control Matrix](/action-control-matrix/) - [All glossary terms](/glossary/) ## Sources - www.island.io/blog/island-the-enterprise-browser-for-ai-agents: https://www.island.io/blog/island-the-enterprise-browser-for-ai-agents (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/glossary/browser-agent/ --- # Shadow AI, defined with a sourced example GLOSSARY, read 12 Nov 2026 **The short answer.** Shadow AI is the use of AI tools that security and IT have not approved or cannot see. Developers using personal accounts for coding assistants are the common case. Knostic defines shadow AI as the unsanctioned use of AI tools by developers outside official oversight, and says developers may paste sensitive code or credentials into external AI systems. ## Definition Shadow AI is the use of AI tools that security and IT have not approved or cannot see. Developers using personal accounts for coding assistants are the common case. The risks are data leakage, unreviewed dependencies and no record of what the tool did. ## How do you detect shadow AI? Combine signals: process and binary inventory, browser extension inventory, network and DNS records, agent and MCP config files, and OAuth grants. Each sees a slice. ## Example from a real source **Example.** Knostic defines shadow AI as the unsanctioned use of AI tools by developers outside official oversight, and says developers may paste sensitive code or credentials into external AI systems. ## Related terms - [AI DLP](/glossary/ai-dlp/): AI DLP is data loss prevention aimed at prompts, uploads and agent context sent to AI tools. - [Data paste control](/glossary/data-paste-control/): A data paste control is a policy that inspects, warns on or blocks pasting content into AI tools, in a browser or a desktop app. - [Browser agent](/glossary/browser-agent/): A browser agent is an AI agent that navigates web pages, reads content, fills forms and clicks on behalf of a user. ## Where to read more - [The Agent Surface Map](/agent-surface-map/) - [The Action Control Matrix](/action-control-matrix/) - [All glossary terms](/glossary/) ## Sources - www.knostic.ai/blog/shadow-ai-risks: https://www.knostic.ai/blog/shadow-ai-risks (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/glossary/shadow-ai/ --- # AI DLP, defined with a sourced example GLOSSARY, read 12 Nov 2026 **The short answer.** AI DLP is data loss prevention aimed at prompts, uploads and agent context sent to AI tools. Harmonic Security says pattern-matching DLP cannot tell a draft email from a deal memo, and that its product classifies the meaning of the work instead of the shape of the string. ## Definition AI DLP is data loss prevention aimed at prompts, uploads and agent context sent to AI tools. Because prompts are free text, vendors argue that classifying meaning works better than matching fixed patterns. Evaluate false positives on your own traffic. ## Example from a real source **Example.** Harmonic Security says pattern-matching DLP cannot tell a draft email from a deal memo, and that its product classifies the meaning of the work instead of the shape of the string. ## Related terms - [Data paste control](/glossary/data-paste-control/): A data paste control is a policy that inspects, warns on or blocks pasting content into AI tools, in a browser or a desktop app. - [Agentless deployment](/glossary/agentless-deployment/): Agentless deployment means coverage that does not install a new persistent sensor on each device. - [Shadow AI](/glossary/shadow-ai/): Shadow AI is the use of AI tools that security and IT have not approved or cannot see. ## Where to read more - [The Agent Surface Map](/agent-surface-map/) - [The Action Control Matrix](/action-control-matrix/) - [All glossary terms](/glossary/) ## Sources - www.harmonic.security/: https://www.harmonic.security/ (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/glossary/ai-dlp/ --- # Data paste control, defined with a sourced example GLOSSARY, read 12 Nov 2026 **The short answer.** A data paste control is a policy that inspects, warns on or blocks pasting content into AI tools, in a browser or a desktop app. Knostic's shadow AI guide lists developers pasting sensitive code or credentials into external AI systems as a top data leakage concern. ## Definition A data paste control is a policy that inspects, warns on or blocks pasting content into AI tools, in a browser or a desktop app. It targets the most common leak path, a person copying source code or credentials into a chat window. ## Example from a real source **Example.** Knostic's shadow AI guide lists developers pasting sensitive code or credentials into external AI systems as a top data leakage concern. ## Related terms - [Agentless deployment](/glossary/agentless-deployment/): Agentless deployment means coverage that does not install a new persistent sensor on each device. - [Endpoint agent](/glossary/endpoint-agent/): An endpoint agent is the software sensor installed on a device to collect telemetry and enforce policy. - [AI DLP](/glossary/ai-dlp/): AI DLP is data loss prevention aimed at prompts, uploads and agent context sent to AI tools. ## Where to read more - [The Agent Surface Map](/agent-surface-map/) - [The Action Control Matrix](/action-control-matrix/) - [All glossary terms](/glossary/) ## Sources - www.knostic.ai/blog/shadow-ai-risks: https://www.knostic.ai/blog/shadow-ai-risks (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/glossary/data-paste-control/ --- # Agentless deployment, defined with a sourced example GLOSSARY, read 12 Nov 2026 **The short answer.** Agentless deployment means coverage that does not install a new persistent sensor on each device. It can speed rollout, though it changes what is visible locally. Bay says its platform brings governance and enforcement across employee devices, cloud and SaaS without an additional persistent endpoint agent, kernel driver or network proxy. ## Definition Agentless deployment means coverage that does not install a new persistent sensor on each device. It can speed rollout, though it changes what is visible locally. Ask a vendor which events it can see and block without a local component, and what remains when a device is offline. ## Example from a real source **Example.** Bay says its platform brings governance and enforcement across employee devices, cloud and SaaS without an additional persistent endpoint agent, kernel driver or network proxy. ## Related terms - [Endpoint agent](/glossary/endpoint-agent/): An endpoint agent is the software sensor installed on a device to collect telemetry and enforce policy. - [Managed settings](/glossary/managed-settings/): Managed settings are configuration that an organization deploys to devices and that users cannot override. - [Data paste control](/glossary/data-paste-control/): A data paste control is a policy that inspects, warns on or blocks pasting content into AI tools, in a browser or a desktop app. ## Where to read more - [The Agent Surface Map](/agent-surface-map/) - [The Action Control Matrix](/action-control-matrix/) - [All glossary terms](/glossary/) ## Sources - www.bay.io/blog/bay-emerges-from-stealth-with-50m-after-fortune-100-companies-used-its-platform-to-triple-autonomous-ai-agent-use-2: https://www.bay.io/blog/bay-emerges-from-stealth-with-50m-after-fortune-100-companies-used-its-platform-to-triple-autonomous-ai-agent-use-2 (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/glossary/agentless-deployment/ --- # Endpoint agent, defined with a sourced example GLOSSARY, read 12 Nov 2026 **The short answer.** An endpoint agent is the software sensor installed on a device to collect telemetry and enforce policy. In AI endpoint security it is the component that sees agent processes, hooks and local network traffic. CrowdStrike says Falcon Guardian unifies agent-level controls, endpoint network inspection and the endpoint on a single Falcon sensor. ## What is an endpoint agent? An endpoint agent is the software sensor installed on a device to collect telemetry and enforce policy. In AI endpoint security it is the component that sees agent processes, hooks and local network traffic. It adds a deployed component to manage, in return for local depth. ## Example from a real source **Example.** CrowdStrike says Falcon Guardian unifies agent-level controls, endpoint network inspection and the endpoint on a single Falcon sensor. ## Related terms - [Managed settings](/glossary/managed-settings/): Managed settings are configuration that an organization deploys to devices and that users cannot override. - [Permission mode](/glossary/permission-mode/): A permission mode is a setting that decides how often an agent must ask before acting, from asking on every action to skipping prompts. - [Agentless deployment](/glossary/agentless-deployment/): Agentless deployment means coverage that does not install a new persistent sensor on each device. ## Where to read more - [The Agent Surface Map](/agent-surface-map/) - [The Action Control Matrix](/action-control-matrix/) - [All glossary terms](/glossary/) ## Sources - www.crowdstrike.com/en-us/press-releases/crowdstrike-delivers-ai-security-safety-for-the-enterprise/: https://www.crowdstrike.com/en-us/press-releases/crowdstrike-delivers-ai-security-safety-for-the-enterprise/ (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/glossary/endpoint-agent/ --- # Managed settings, defined with a sourced example GLOSSARY, read 12 Nov 2026 **The short answer.** Managed settings are configuration that an organization deploys to devices and that users cannot override. For coding agents they set permission rules, disable risky modes and restrict plugins or servers. Anthropic's documentation says managed settings sit at the top of Claude Code's precedence, and nothing in a user's own settings files overrides a managed key. ## Definition Managed settings are configuration that an organization deploys to devices and that users cannot override. For coding agents they set permission rules, disable risky modes and restrict plugins or servers. They are the main way to apply policy to a tool that has no central console of its own. ## Example from a real source **Example.** Anthropic's documentation says managed settings sit at the top of Claude Code's precedence, and nothing in a user's own settings files overrides a managed key. ## Related terms - [Permission mode](/glossary/permission-mode/): A permission mode is a setting that decides how often an agent must ask before acting, from asking on every action to skipping prompts. - [Allowlist](/glossary/allowlist/): An allowlist is a documented list of elements that policy permits, with everything else denied. - [Endpoint agent](/glossary/endpoint-agent/): An endpoint agent is the software sensor installed on a device to collect telemetry and enforce policy. ## Where to read more - [The Agent Surface Map](/agent-surface-map/) - [The Action Control Matrix](/action-control-matrix/) - [All glossary terms](/glossary/) ## Sources - code.claude.com/docs/en/settings: https://code.claude.com/docs/en/settings (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/glossary/managed-settings/ --- # Permission mode, defined with a sourced example GLOSSARY, read 12 Nov 2026 **The short answer.** A permission mode is a setting that decides how often an agent must ask before acting, from asking on every action to skipping prompts. Anthropic's permissions documentation lists modes for Claude Code, including bypassPermissions, and notes that policy can disable that mode and auto mode in settings. ## Definition A permission mode is a setting that decides how often an agent must ask before acting, from asking on every action to skipping prompts. The mode a developer picks changes the real risk of the same tool, so it belongs in policy. ## Example from a real source **Example.** Anthropic's permissions documentation lists modes for Claude Code, including bypassPermissions, and notes that policy can disable that mode and auto mode in settings. ## Related terms - [Allowlist](/glossary/allowlist/): An allowlist is a documented list of elements that policy permits, with everything else denied. - [Sandbox](/glossary/sandbox/): A sandbox is a restricted, controlled execution environment that limits what software can reach, typically the file system and network. - [Managed settings](/glossary/managed-settings/): Managed settings are configuration that an organization deploys to devices and that users cannot override. ## Where to read more - [The Agent Surface Map](/agent-surface-map/) - [The Action Control Matrix](/action-control-matrix/) - [All glossary terms](/glossary/) ## Sources - code.claude.com/docs/en/permissions: https://code.claude.com/docs/en/permissions (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/glossary/permission-mode/ --- # Allowlist, defined with a sourced example GLOSSARY, read 12 Nov 2026 **The short answer.** An allowlist is a documented list of elements that policy permits, with everything else denied. Cursor's agent security documentation says trusted tool calls can run without prompting through run modes that range from a simple allowlist to an auto-review classifier, and calls them best-effort guardrails rather than a hard security boundary. ## Definition An allowlist is a documented list of elements that policy permits, with everything else denied. For agents it can cover commands, domains, MCP servers, extensions or plugins. It is stricter than a blocklist because new, unknown items are denied by default. ## Example from a real source **Example.** Cursor's agent security documentation says trusted tool calls can run without prompting through run modes that range from a simple allowlist to an auto-review classifier, and calls them best-effort guardrails rather than a hard security boundary. ## Related terms - [Sandbox](/glossary/sandbox/): A sandbox is a restricted, controlled execution environment that limits what software can reach, typically the file system and network. - [Prompt injection](/glossary/prompt-injection/): Prompt injection is when input to a language model changes its behavior in ways its operator did not intend. - [Permission mode](/glossary/permission-mode/): A permission mode is a setting that decides how often an agent must ask before acting, from asking on every action to skipping prompts. ## Where to read more - [The Agent Surface Map](/agent-surface-map/) - [The Action Control Matrix](/action-control-matrix/) - [All glossary terms](/glossary/) ## Sources - cursor.com/docs/agent/security: https://cursor.com/docs/agent/security (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/glossary/allowlist/ --- # Sandbox, defined with a sourced example GLOSSARY, read 12 Nov 2026 **The short answer.** A sandbox is a restricted, controlled execution environment that limits what software can reach, typically the file system and network. For coding agents it bounds what a command can touch even if the agent is misled. Google's Gemini CLI documentation says sandboxing isolates potentially dangerous operations such as shell commands or file modifications from the host system. ## Definition A sandbox is a restricted, controlled execution environment that limits what software can reach, typically the file system and network. For coding agents it bounds what a command can touch even if the agent is misled. Sandbox flaws are therefore security bugs. ## Example from a real source **Example.** Google's Gemini CLI documentation says sandboxing isolates potentially dangerous operations such as shell commands or file modifications from the host system. ## Related terms - [Prompt injection](/glossary/prompt-injection/): Prompt injection is when input to a language model changes its behavior in ways its operator did not intend. - [Indirect prompt injection](/glossary/indirect-prompt-injection/): Indirect prompt injection happens when the instructions come from content the agent reads, such as a web page, a file or a repository, instead of from the user. - [Allowlist](/glossary/allowlist/): An allowlist is a documented list of elements that policy permits, with everything else denied. ## Where to read more - [The Agent Surface Map](/agent-surface-map/) - [The Action Control Matrix](/action-control-matrix/) - [All glossary terms](/glossary/) ## Sources - geminicli.com/docs/cli/sandbox/: https://geminicli.com/docs/cli/sandbox/ (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/glossary/sandbox/ --- # Prompt injection, defined with a sourced example GLOSSARY, read 12 Nov 2026 **The short answer.** Prompt injection is when input to a language model changes its behavior in ways its operator did not intend. OWASP's LLM01:2025 entry says the inputs can affect the model even if humans cannot see or read them, as long as the model parses the content. ## Definition Prompt injection is when input to a language model changes its behavior in ways its operator did not intend. In an agent with tools, the changed behavior can be an action, such as reading a file or sending a request. OWASP lists it as the first risk for LLM applications. ## Example from a real source **Example.** OWASP's LLM01:2025 entry says the inputs can affect the model even if humans cannot see or read them, as long as the model parses the content. ## Related terms - [Indirect prompt injection](/glossary/indirect-prompt-injection/): Indirect prompt injection happens when the instructions come from content the agent reads, such as a web page, a file or a repository, instead of from the user. - [Tool poisoning](/glossary/tool-poisoning/): Tool poisoning is hiding instructions in an MCP tool's description or metadata. - [Sandbox](/glossary/sandbox/): A sandbox is a restricted, controlled execution environment that limits what software can reach, typically the file system and network. ## Where to read more - [The Agent Surface Map](/agent-surface-map/) - [The Action Control Matrix](/action-control-matrix/) - [All glossary terms](/glossary/) ## Sources - genai.owasp.org/llmrisk/llm01-prompt-injection/: https://genai.owasp.org/llmrisk/llm01-prompt-injection/ (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/glossary/prompt-injection/ --- # Indirect prompt injection, defined with a sourced example GLOSSARY, read 12 Nov 2026 **The short answer.** Indirect prompt injection happens when the instructions come from content the agent reads, such as a web page, a file or a repository, instead of from the user. OWASP says indirect prompt injections occur when an LLM accepts input from external sources such as websites or files. ## Definition Indirect prompt injection happens when the instructions come from content the agent reads, such as a web page, a file or a repository, instead of from the user. The agent cannot reliably tell data from commands, so any untrusted content it ingests is a possible channel. ## Example from a real source **Example.** OWASP says indirect prompt injections occur when an LLM accepts input from external sources such as websites or files. ## Related terms - [Tool poisoning](/glossary/tool-poisoning/): Tool poisoning is hiding instructions in an MCP tool's description or metadata. - [Slopsquatting / package hallucination](/glossary/slopsquatting-package-hallucination/): Package hallucination is when a model suggests a software package name that does not exist. - [Prompt injection](/glossary/prompt-injection/): Prompt injection is when input to a language model changes its behavior in ways its operator did not intend. ## Where to read more - [The Agent Surface Map](/agent-surface-map/) - [The Action Control Matrix](/action-control-matrix/) - [All glossary terms](/glossary/) ## Sources - genai.owasp.org/llmrisk/llm01-prompt-injection/: https://genai.owasp.org/llmrisk/llm01-prompt-injection/ (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/glossary/indirect-prompt-injection/ --- # Tool poisoning, defined with a sourced example GLOSSARY, read 12 Nov 2026 **The short answer.** Tool poisoning is hiding instructions in an MCP tool's description or metadata. The model reads them and may act on them, while the user usually sees only a short summary. Invariant Labs described a Tool Poisoning Attack as malicious instructions embedded in MCP tool descriptions that are invisible to users but visible to AI models. ## What is MCP tool poisoning? Tool poisoning is hiding instructions in an MCP tool's description or metadata. The model reads them and may act on them, while the user usually sees only a short summary. Review tool descriptions, not only tool names, when you vet a server. ## Example from a real source **Example.** Invariant Labs described a Tool Poisoning Attack as malicious instructions embedded in MCP tool descriptions that are invisible to users but visible to AI models. ## Related terms - [Slopsquatting / package hallucination](/glossary/slopsquatting-package-hallucination/): Package hallucination is when a model suggests a software package name that does not exist. - [Credential exposure](/glossary/credential-exposure/): Credential exposure is when tokens, keys or passwords become readable to an agent, a tool or an attacker. - [Indirect prompt injection](/glossary/indirect-prompt-injection/): Indirect prompt injection happens when the instructions come from content the agent reads, such as a web page, a file or a repository, instead of from the user. ## Where to read more - [The Agent Surface Map](/agent-surface-map/) - [The Action Control Matrix](/action-control-matrix/) - [All glossary terms](/glossary/) ## Sources - invariantlabs.ai/blog/mcp-security-notification-tool-poisoning-attacks: https://invariantlabs.ai/blog/mcp-security-notification-tool-poisoning-attacks (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/glossary/tool-poisoning/ --- # Slopsquatting / package hallucination, defined with a sourced example GLOSSARY, read 12 Nov 2026 **The short answer.** Package hallucination is when a model suggests a software package name that does not exist. Wikipedia defines slopsquatting as registering a non-existent package name that a large language model may hallucinate, and notes that researcher Bar Lanyado saw a model invent the name huggingface-cli in 2023. ## Definition Package hallucination is when a model suggests a software package name that does not exist. Slopsquatting is an attacker registering that name so a developer or agent who installs it gets malicious code. Pin dependencies and check that a suggested package is real and established before install. ## Example from a real source **Example.** Wikipedia defines slopsquatting as registering a non-existent package name that a large language model may hallucinate, and notes that researcher Bar Lanyado saw a model invent the name huggingface-cli in 2023. ## Related terms - [Credential exposure](/glossary/credential-exposure/): Credential exposure is when tokens, keys or passwords become readable to an agent, a tool or an attacker. - [Secrets in context](/glossary/secrets-in-context/): Secrets in context are credentials or private data that have been loaded into an agent's context window. - [Tool poisoning](/glossary/tool-poisoning/): Tool poisoning is hiding instructions in an MCP tool's description or metadata. ## Where to read more - [The Agent Surface Map](/agent-surface-map/) - [The Action Control Matrix](/action-control-matrix/) - [All glossary terms](/glossary/) ## Sources - en.wikipedia.org/wiki/Slopsquatting: https://en.wikipedia.org/wiki/Slopsquatting (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/glossary/slopsquatting-package-hallucination/ --- # Credential exposure, defined with a sourced example GLOSSARY, read 12 Nov 2026 **The short answer.** Credential exposure is when tokens, keys or passwords become readable to an agent, a tool or an attacker. Agents can ingest them from files, environment variables, IDE state and cloud profiles. Short-lived, scoped tokens limit the damage. Knostic says credential sprawl happens when coding agents automatically ingest sensitive tokens from files, APIs, IDEs and cloud environments without explicit user action. ## Definition Credential exposure is when tokens, keys or passwords become readable to an agent, a tool or an attacker. Agents can ingest them from files, environment variables, IDE state and cloud profiles. Short-lived, scoped tokens limit the damage. ## Example from a real source **Example.** Knostic says credential sprawl happens when coding agents automatically ingest sensitive tokens from files, APIs, IDEs and cloud environments without explicit user action. ## Related terms - [Secrets in context](/glossary/secrets-in-context/): Secrets in context are credentials or private data that have been loaded into an agent's context window. - [EDR](/glossary/edr/): Endpoint detection and response is a security tool that continuously monitors devices, records system-level behavior, detects suspicious activity and helps respond. - [Slopsquatting / package hallucination](/glossary/slopsquatting-package-hallucination/): Package hallucination is when a model suggests a software package name that does not exist. ## Where to read more - [The Agent Surface Map](/agent-surface-map/) - [The Action Control Matrix](/action-control-matrix/) - [All glossary terms](/glossary/) ## Sources - www.knostic.ai/blog/credential-management-coding-agents: https://www.knostic.ai/blog/credential-management-coding-agents (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/glossary/credential-exposure/ --- # Secrets in context, defined with a sourced example GLOSSARY, read 12 Nov 2026 **The short answer.** Secrets in context are credentials or private data that have been loaded into an agent's context window. OWASP's LLM02:2025 entry on leaks of sensitive information lists security credentials among the sensitive data that can affect both the model and its application context. ## Definition Secrets in context are credentials or private data that have been loaded into an agent's context window. Anything in context can be sent to a model provider, echoed in output or passed to a tool call. Keep secrets out of files and prompts an agent reads. ## Example from a real source **Example.** OWASP's LLM02:2025 entry on leaks of sensitive information lists security credentials among the sensitive data that can affect both the model and its application context. ## Related terms - [EDR](/glossary/edr/): Endpoint detection and response is a security tool that continuously monitors devices, records system-level behavior, detects suspicious activity and helps respond. - [Audit trail](/glossary/audit-trail/): An audit trail is a chronological record that lets you reconstruct the sequence of activities around an event. - [Credential exposure](/glossary/credential-exposure/): Credential exposure is when tokens, keys or passwords become readable to an agent, a tool or an attacker. ## Where to read more - [The Agent Surface Map](/agent-surface-map/) - [The Action Control Matrix](/action-control-matrix/) - [All glossary terms](/glossary/) ## Sources - genai.owasp.org/llmrisk/llm022025-sensitive-information-disclosure/: https://genai.owasp.org/llmrisk/llm022025-sensitive-information-disclosure/ (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/glossary/secrets-in-context/ --- # EDR, defined with a sourced example GLOSSARY, read 12 Nov 2026 **The short answer.** Endpoint detection and response is a security tool that continuously monitors devices, records system-level behavior, detects suspicious activity and helps respond. CrowdStrike defines EDR as an endpoint security solution that continuously monitors end-user devices to detect and respond to threats such as ransomware and malware. ## Definition Endpoint detection and response is a security tool that continuously monitors devices, records system-level behavior, detects suspicious activity and helps respond. It watches processes and files, so an approved agent doing something harmful with valid access can look normal to it. ## What is the difference between AI endpoint security and EDR? EDR records and responds to process, file and network behavior on a device. It generally cannot tell a shell command run by a coding agent from one typed by a developer, and it does not read prompts or tool calls. ## Example from a real source **Example.** CrowdStrike defines EDR as an endpoint security solution that continuously monitors end-user devices to detect and respond to threats such as ransomware and malware. ## Related terms - [Audit trail](/glossary/audit-trail/): An audit trail is a chronological record that lets you reconstruct the sequence of activities around an event. - [Runtime guardrail](/glossary/runtime-guardrail/): A runtime guardrail is a control that inspects and can block an agent's prompts or actions while it runs, as opposed to scanning it before deployment. - [Secrets in context](/glossary/secrets-in-context/): Secrets in context are credentials or private data that have been loaded into an agent's context window. ## Where to read more - [The Agent Surface Map](/agent-surface-map/) - [The Action Control Matrix](/action-control-matrix/) - [All glossary terms](/glossary/) ## Sources - www.crowdstrike.com/en-us/cybersecurity-101/endpoint-security/endpoint-detection-and-response-edr/: https://www.crowdstrike.com/en-us/cybersecurity-101/endpoint-security/endpoint-detection-and-response-edr/ (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/glossary/edr/ --- # Audit trail, defined with a sourced example GLOSSARY, read 12 Nov 2026 **The short answer.** An audit trail is a chronological record that lets you reconstruct the sequence of activities around an event. NIST's glossary defines an audit trail as a chronological record that reconstructs and examines the sequence of activities surrounding a security-relevant transaction, citing NIST SP 800-53 Rev. 5. ## Definition An audit trail is a chronological record that lets you reconstruct the sequence of activities around an event. For agents it means prompts, tool calls, file changes and approvals tied to a user and a session. Without it, an incident review has nothing to start from. ## Example from a real source **Example.** NIST's glossary defines an audit trail as a chronological record that reconstructs and examines the sequence of activities surrounding a security-relevant transaction, citing NIST SP 800-53 Rev. 5. ## Related terms - [Runtime guardrail](/glossary/runtime-guardrail/): A runtime guardrail is a control that inspects and can block an agent's prompts or actions while it runs, as opposed to scanning it before deployment. - [Pre-execution block](/glossary/pre-execution-block/): A pre-execution block stops a tool call before it runs, based on policy, instead of detecting it afterward. - [EDR](/glossary/edr/): Endpoint detection and response is a security tool that continuously monitors devices, records system-level behavior, detects suspicious activity and helps respond. ## Where to read more - [The Agent Surface Map](/agent-surface-map/) - [The Action Control Matrix](/action-control-matrix/) - [All glossary terms](/glossary/) ## Sources - csrc.nist.gov/glossary/term/audit_trail: https://csrc.nist.gov/glossary/term/audit_trail (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/glossary/audit-trail/ --- # Runtime guardrail, defined with a sourced example GLOSSARY, read 12 Nov 2026 **The short answer.** A runtime guardrail is a control that inspects and can block an agent's prompts or actions while it runs, as opposed to scanning it before deployment. CrowdStrike says its agent-level hooks enforce policy inside Claude Code, OpenAI Codex, GitHub Copilot, Microsoft 365 Copilot, OpenCode and Cursor, to inspect prompts and block malicious activity at the source. ## Definition A runtime guardrail is a control that inspects and can block an agent's prompts or actions while it runs, as opposed to scanning it before deployment. It might sit in a hook, a gateway or an endpoint sensor. ## Example from a real source **Example.** CrowdStrike says its agent-level hooks enforce policy inside Claude Code, OpenAI Codex, GitHub Copilot, Microsoft 365 Copilot, OpenCode and Cursor, to inspect prompts and block malicious activity at the source. ## Related terms - [Pre-execution block](/glossary/pre-execution-block/): A pre-execution block stops a tool call before it runs, based on policy, instead of detecting it afterward. - [Inventory](/glossary/inventory/): An inventory is a current list of the AI tools, agents, MCP servers, skills and extensions on each device, with versions and owners. - [Audit trail](/glossary/audit-trail/): An audit trail is a chronological record that lets you reconstruct the sequence of activities around an event. ## Where to read more - [The Agent Surface Map](/agent-surface-map/) - [The Action Control Matrix](/action-control-matrix/) - [All glossary terms](/glossary/) ## Sources - www.crowdstrike.com/en-us/press-releases/crowdstrike-delivers-ai-security-safety-for-the-enterprise/: https://www.crowdstrike.com/en-us/press-releases/crowdstrike-delivers-ai-security-safety-for-the-enterprise/ (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/glossary/runtime-guardrail/ --- # Pre-execution block, defined with a sourced example GLOSSARY, read 12 Nov 2026 **The short answer.** A pre-execution block stops a tool call before it runs, based on policy, instead of detecting it afterward. It needs a hook point ahead of the action. Compare it with after-the-fact alerting when you judge a product. Anthropic's hooks reference documents a PreToolUse event with decision control, which runs before a tool call and can change whether it proceeds. ## Definition A pre-execution block stops a tool call before it runs, based on policy, instead of detecting it afterward. It needs a hook point ahead of the action. Compare it with after-the-fact alerting when you judge a product. ## Example from a real source **Example.** Anthropic's hooks reference documents a PreToolUse event with decision control, which runs before a tool call and can change whether it proceeds. ## Related terms - [Inventory](/glossary/inventory/): An inventory is a current list of the AI tools, agents, MCP servers, skills and extensions on each device, with versions and owners. - [Desktop AI app](/glossary/desktop-ai-app/): A desktop AI app is a native application, such as an AI chat client with local file access, that runs on the user's machine. - [Runtime guardrail](/glossary/runtime-guardrail/): A runtime guardrail is a control that inspects and can block an agent's prompts or actions while it runs, as opposed to scanning it before deployment. ## Where to read more - [The Agent Surface Map](/agent-surface-map/) - [The Action Control Matrix](/action-control-matrix/) - [All glossary terms](/glossary/) ## Sources - code.claude.com/docs/en/hooks: https://code.claude.com/docs/en/hooks (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/glossary/pre-execution-block/ --- # Inventory, defined with a sourced example GLOSSARY, read 12 Nov 2026 **The short answer.** An inventory is a current list of the AI tools, agents, MCP servers, skills and extensions on each device, with versions and owners. Lasso says an endpoint program starts with knowing which AI tools and agents exist on a device, then controlling their access, protecting data and monitoring behavior. ## Definition An inventory is a current list of the AI tools, agents, MCP servers, skills and extensions on each device, with versions and owners. It is the base for every other control, because you cannot apply policy to what you cannot list. ## Example from a real source **Example.** Lasso says an endpoint program starts with knowing which AI tools and agents exist on a device, then controlling their access, protecting data and monitoring behavior. ## Related terms - [Desktop AI app](/glossary/desktop-ai-app/): A desktop AI app is a native application, such as an AI chat client with local file access, that runs on the user's machine. - [IDE extension marketplace](/glossary/ide-extension-marketplace/): An IDE extension marketplace is the catalog where editors such as VS Code and Cursor fetch extensions. - [Pre-execution block](/glossary/pre-execution-block/): A pre-execution block stops a tool call before it runs, based on policy, instead of detecting it afterward. ## Where to read more - [The Agent Surface Map](/agent-surface-map/) - [The Action Control Matrix](/action-control-matrix/) - [All glossary terms](/glossary/) ## Sources - www.lasso.security/blog/securing-ai-on-endpoints-how-to-govern-ai-tools-coding-agents-and-mcp-connections: https://www.lasso.security/blog/securing-ai-on-endpoints-how-to-govern-ai-tools-coding-agents-and-mcp-connections (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/glossary/inventory/ --- # Desktop AI app, defined with a sourced example GLOSSARY, read 12 Nov 2026 **The short answer.** A desktop AI app is a native application, such as an AI chat client with local file access, that runs on the user's machine. Anthropic's advisory GHSA-v234-4jrq-mgg6 describes Claude Desktop on macOS and a block list of file types that Cowork sessions may not open, fixed in Claude Desktop 1.15962.0. ## Definition A desktop AI app is a native application, such as an AI chat client with local file access, that runs on the user's machine. It has its own update channel, local features and attack surface, separate from the browser. ## Example from a real source **Example.** Anthropic's advisory GHSA-v234-4jrq-mgg6 describes Claude Desktop on macOS and a block list of file types that Cowork sessions may not open, fixed in Claude Desktop 1.15962.0. ## Related terms - [IDE extension marketplace](/glossary/ide-extension-marketplace/): An IDE extension marketplace is the catalog where editors such as VS Code and Cursor fetch extensions. - [Rug pull (MCP)](/glossary/rug-pull-mcp/): In MCP, a rug pull is when a server changes a tool's description or behavior after the user approved it. - [Inventory](/glossary/inventory/): An inventory is a current list of the AI tools, agents, MCP servers, skills and extensions on each device, with versions and owners. ## Where to read more - [The Agent Surface Map](/agent-surface-map/) - [The Action Control Matrix](/action-control-matrix/) - [All glossary terms](/glossary/) ## Sources - github.com/anthropics/claude-code/security/advisories/GHSA-v234-4jrq-mgg6: https://github.com/anthropics/claude-code/security/advisories/GHSA-v234-4jrq-mgg6 (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/glossary/desktop-ai-app/ --- # IDE extension marketplace, defined with a sourced example GLOSSARY, read 12 Nov 2026 **The short answer.** An IDE extension marketplace is the catalog where editors such as VS Code and Cursor fetch extensions. Extensions run inside the editor with the developer's rights, so a bad or tampered listing can reach tokens and repositories. Noma Security reported that a backdoored build of the anthropic.claude-code extension, version 2.1.313, was distributed through Cursor's extension marketplace before being pulled. ## Definition An IDE extension marketplace is the catalog where editors such as VS Code and Cursor fetch extensions. Extensions run inside the editor with the developer's rights, so a bad or tampered listing can reach tokens and repositories. Many AI coding tools ship as extensions. ## Example from a real source **Example.** Noma Security reported that a backdoored build of the anthropic.claude-code extension, version 2.1.313, was distributed through Cursor's extension marketplace before being pulled. ## Related terms - [Rug pull (MCP)](/glossary/rug-pull-mcp/): In MCP, a rug pull is when a server changes a tool's description or behavior after the user approved it. - [Least privilege for agents](/glossary/least-privilege-for-agents/): Least privilege for agents means giving an agent only the files, tools, network access and credentials its task needs, for as long as it needs them. - [Desktop AI app](/glossary/desktop-ai-app/): A desktop AI app is a native application, such as an AI chat client with local file access, that runs on the user's machine. ## Where to read more - [The Agent Surface Map](/agent-surface-map/) - [The Action Control Matrix](/action-control-matrix/) - [All glossary terms](/glossary/) ## Sources - noma.security/blog/supply-chain-attack-poisoned-claude-code-extension-installs-hidden-backdoors: https://noma.security/blog/supply-chain-attack-poisoned-claude-code-extension-installs-hidden-backdoors (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/glossary/ide-extension-marketplace/ --- # Rug pull (MCP), defined with a sourced example GLOSSARY, read 12 Nov 2026 **The short answer.** In MCP, a rug pull is when a server changes a tool's description or behavior after the user approved it. The approval then covers something different from what the user saw. Pin versions and re-review when a server updates. Invariant Labs defined it as a malicious server changing the tool description after the client has already approved it. ## Definition In MCP, a rug pull is when a server changes a tool's description or behavior after the user approved it. The approval then covers something different from what the user saw. Pin versions and re-review when a server updates. ## Example from a real source **Example.** Invariant Labs defined it as a malicious server changing the tool description after the client has already approved it. ## Related terms - [Least privilege for agents](/glossary/least-privilege-for-agents/): Least privilege for agents means giving an agent only the files, tools, network access and credentials its task needs, for as long as it needs them. - [Human-in-the-loop approval](/glossary/human-in-the-loop-approval/): Human-in-the-loop approval requires a person to confirm an agent action before it runs. - [IDE extension marketplace](/glossary/ide-extension-marketplace/): An IDE extension marketplace is the catalog where editors such as VS Code and Cursor fetch extensions. ## Where to read more - [The Agent Surface Map](/agent-surface-map/) - [The Action Control Matrix](/action-control-matrix/) - [All glossary terms](/glossary/) ## Sources - invariantlabs.ai/blog/mcp-security-notification-tool-poisoning-attacks: https://invariantlabs.ai/blog/mcp-security-notification-tool-poisoning-attacks (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/glossary/rug-pull-mcp/ --- # Least privilege for agents, defined with a sourced example GLOSSARY, read 12 Nov 2026 **The short answer.** Least privilege for agents means giving an agent only the files, tools, network access and credentials its task needs, for as long as it needs them. NIST's glossary describes least privilege as restricting access of users, or processes acting for them, to the minimum necessary to accomplish assigned tasks. ## Definition Least privilege for agents means giving an agent only the files, tools, network access and credentials its task needs, for as long as it needs them. It limits the harm when an agent is misled, since the agent inherits whatever the user holds. ## Example from a real source **Example.** NIST's glossary describes least privilege as restricting access of users, or processes acting for them, to the minimum necessary to accomplish assigned tasks. ## Related terms - [Human-in-the-loop approval](/glossary/human-in-the-loop-approval/): Human-in-the-loop approval requires a person to confirm an agent action before it runs. - [YOLO / auto-approve mode](/glossary/yolo-auto-approve-mode/): Auto-approve mode, often called YOLO mode, lets an agent run actions without asking each time. - [Rug pull (MCP)](/glossary/rug-pull-mcp/): In MCP, a rug pull is when a server changes a tool's description or behavior after the user approved it. ## Where to read more - [The Agent Surface Map](/agent-surface-map/) - [The Action Control Matrix](/action-control-matrix/) - [All glossary terms](/glossary/) ## Sources - csrc.nist.gov/glossary/term/least_privilege: https://csrc.nist.gov/glossary/term/least_privilege (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/glossary/least-privilege-for-agents/ --- # Human-in-the-loop approval, defined with a sourced example GLOSSARY, read 12 Nov 2026 **The short answer.** Human-in-the-loop approval requires a person to confirm an agent action before it runs. The MCP tools specification says there SHOULD always be a human in the loop with the ability to deny tool invocations, and that applications should show confirmation prompts. ## Definition Human-in-the-loop approval requires a person to confirm an agent action before it runs. It works only if the prompt shows what will happen, and breaks down when users approve in bulk or for a whole session. ## Example from a real source **Example.** The MCP tools specification says there SHOULD always be a human in the loop with the ability to deny tool invocations, and that applications should show confirmation prompts. ## Related terms - [YOLO / auto-approve mode](/glossary/yolo-auto-approve-mode/): Auto-approve mode, often called YOLO mode, lets an agent run actions without asking each time. - [AI endpoint security](/glossary/ai-endpoint-security/): AI endpoint security is the practice of finding, governing and stopping the AI tools that run on employee devices. - [Least privilege for agents](/glossary/least-privilege-for-agents/): Least privilege for agents means giving an agent only the files, tools, network access and credentials its task needs, for as long as it needs them. ## Where to read more - [The Agent Surface Map](/agent-surface-map/) - [The Action Control Matrix](/action-control-matrix/) - [All glossary terms](/glossary/) ## Sources - modelcontextprotocol.io/specification/2025-06-18/server/tools: https://modelcontextprotocol.io/specification/2025-06-18/server/tools (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/glossary/human-in-the-loop-approval/ --- # YOLO / auto-approve mode, defined with a sourced example GLOSSARY, read 12 Nov 2026 **The short answer.** Auto-approve mode, often called YOLO mode, lets an agent run actions without asking each time. It is fast, and it removes the approval step that catches mistakes and injected instructions. Anthropic's documentation says in bypassPermissions mode Claude Code skips permission prompts and should only be used in isolated environments such as containers or VMs. ## Definition Auto-approve mode, often called YOLO mode, lets an agent run actions without asking each time. It is fast, and it removes the approval step that catches mistakes and injected instructions. Reserve it for isolated machines and block it by policy elsewhere. ## Example from a real source **Example.** Anthropic's documentation says in bypassPermissions mode Claude Code skips permission prompts and should only be used in isolated environments such as containers or VMs. ## Related terms - [AI endpoint security](/glossary/ai-endpoint-security/): AI endpoint security is the practice of finding, governing and stopping the AI tools that run on employee devices. - [Agentic endpoint](/glossary/agentic-endpoint/): An agentic endpoint is a laptop, workstation or developer machine where autonomous AI agents act with the user's access. - [Human-in-the-loop approval](/glossary/human-in-the-loop-approval/): Human-in-the-loop approval requires a person to confirm an agent action before it runs. ## Where to read more - [The Agent Surface Map](/agent-surface-map/) - [The Action Control Matrix](/action-control-matrix/) - [All glossary terms](/glossary/) ## Sources - code.claude.com/docs/en/permissions: https://code.claude.com/docs/en/permissions (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/glossary/yolo-auto-approve-mode/ --- # 36 terms for AI endpoint security, defined here with a sourced example on each term page GLOSSARY, read 12 Nov 2026 **The short answer.** The glossary defines 36 terms used across the index, from agent inventory and shadow AI to MCP servers, prompt injection and endpoint detection and response. Each term has a plain definition and an example taken from a primary source, with the URL and the date it was read. - [AI endpoint security](/glossary/ai-endpoint-security/): AI endpoint security is the practice of finding, governing and stopping the AI tools that run on employee devices. - [Agentic endpoint](/glossary/agentic-endpoint/): An agentic endpoint is a laptop, workstation or developer machine where autonomous AI agents act with the user's access. - [Coding agent](/glossary/coding-agent/): A coding agent is an AI tool that reads a repository, edits files, runs commands and calls tools to finish a development task with limited step-by-step direction. - [Command-line agent](/glossary/command-line-agent/): A command-line agent is a coding agent that runs in a terminal session instead of an editor window. - [MCP server (local stdio)](/glossary/mcp-server-local-stdio/): A local stdio MCP server is a program the MCP client launches as a subprocess on the same machine, talking over standard input and output. - [MCP client](/glossary/mcp-client/): An MCP client is the connector inside an AI application that holds a dedicated connection to one MCP server. - [Tool call](/glossary/tool-call/): A tool call is a request from the model, run through the client, to execute a named function on a server or in the agent itself, with arguments. - [Agent skill](/glossary/agent-skill/): An agent skill is a packaged set of instructions, and sometimes scripts, that an agent loads when a task matches. - [Plugin](/glossary/plugin/): In a coding agent, a plugin is a bundle that can contain skills, sub-agents, hooks and MCP servers, installed as one unit, usually from a marketplace. - [Browser agent](/glossary/browser-agent/): A browser agent is an AI agent that navigates web pages, reads content, fills forms and clicks on behalf of a user. - [Shadow AI](/glossary/shadow-ai/): Shadow AI is the use of AI tools that security and IT have not approved or cannot see. - [AI DLP](/glossary/ai-dlp/): AI DLP is data loss prevention aimed at prompts, uploads and agent context sent to AI tools. - [Data paste control](/glossary/data-paste-control/): A data paste control is a policy that inspects, warns on or blocks pasting content into AI tools, in a browser or a desktop app. - [Agentless deployment](/glossary/agentless-deployment/): Agentless deployment means coverage that does not install a new persistent sensor on each device. - [Endpoint agent](/glossary/endpoint-agent/): An endpoint agent is the software sensor installed on a device to collect telemetry and enforce policy. - [Managed settings](/glossary/managed-settings/): Managed settings are configuration that an organization deploys to devices and that users cannot override. - [Permission mode](/glossary/permission-mode/): A permission mode is a setting that decides how often an agent must ask before acting, from asking on every action to skipping prompts. - [Allowlist](/glossary/allowlist/): An allowlist is a documented list of elements that policy permits, with everything else denied. - [Sandbox](/glossary/sandbox/): A sandbox is a restricted, controlled execution environment that limits what software can reach, typically the file system and network. - [Prompt injection](/glossary/prompt-injection/): Prompt injection is when input to a language model changes its behavior in ways its operator did not intend. - [Indirect prompt injection](/glossary/indirect-prompt-injection/): Indirect prompt injection happens when the instructions come from content the agent reads, such as a web page, a file or a repository, instead of from the user. - [Tool poisoning](/glossary/tool-poisoning/): Tool poisoning is hiding instructions in an MCP tool's description or metadata. - [Slopsquatting / package hallucination](/glossary/slopsquatting-package-hallucination/): Package hallucination is when a model suggests a software package name that does not exist. - [Credential exposure](/glossary/credential-exposure/): Credential exposure is when tokens, keys or passwords become readable to an agent, a tool or an attacker. - [Secrets in context](/glossary/secrets-in-context/): Secrets in context are credentials or private data that have been loaded into an agent's context window. - [EDR](/glossary/edr/): Endpoint detection and response is a security tool that continuously monitors devices, records system-level behavior, detects suspicious activity and helps respond. - [Audit trail](/glossary/audit-trail/): An audit trail is a chronological record that lets you reconstruct the sequence of activities around an event. - [Runtime guardrail](/glossary/runtime-guardrail/): A runtime guardrail is a control that inspects and can block an agent's prompts or actions while it runs, as opposed to scanning it before deployment. - [Pre-execution block](/glossary/pre-execution-block/): A pre-execution block stops a tool call before it runs, based on policy, instead of detecting it afterward. - [Inventory](/glossary/inventory/): An inventory is a current list of the AI tools, agents, MCP servers, skills and extensions on each device, with versions and owners. - [Desktop AI app](/glossary/desktop-ai-app/): A desktop AI app is a native application, such as an AI chat client with local file access, that runs on the user's machine. - [IDE extension marketplace](/glossary/ide-extension-marketplace/): An IDE extension marketplace is the catalog where editors such as VS Code and Cursor fetch extensions. - [Rug pull (MCP)](/glossary/rug-pull-mcp/): In MCP, a rug pull is when a server changes a tool's description or behavior after the user approved it. - [Least privilege for agents](/glossary/least-privilege-for-agents/): Least privilege for agents means giving an agent only the files, tools, network access and credentials its task needs, for as long as it needs them. - [Human-in-the-loop approval](/glossary/human-in-the-loop-approval/): Human-in-the-loop approval requires a person to confirm an agent action before it runs. - [YOLO / auto-approve mode](/glossary/yolo-auto-approve-mode/): Auto-approve mode, often called YOLO mode, lets an agent run actions without asking each time. Short answers to 30 common questions are on the [FAQ page](/faq/). ## Sources - Method: https://aiendpointsecurityranked.com/method/ (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/glossary/ --- # VS Code 1.145.3 fixes MCP client and agent host code execution flaws NEWS, 10 NOV 2026, read 12 Nov 2026 **The short answer.** Microsoft published a batch of VS Code advisories on 10 Nov 2026, fixed in 1.145.3. They cover the MCP client (legacy SSE endpoints, elicitation links, install links) and the Agent Host, where agent-originated local commands could skip terminal approval. The MCP elicitation flaw is rated High, CVSS 8.3. On 10 Nov 2026 the VS Code repository published several advisories affecting versions 1.145.2 and earlier, all fixed in 1.145.3. Three involve the MCP client. In the legacy SSE transport, a remote server could supply a follow-up endpoint that VS Code used without validating origin and protocol (GHSA-4x35-6qr8-ww48). A malicious server could also put a clickable command link in an elicitation notification (GHSA-66hq-jw93-462h, High 8.3). A crafted MCP install link could carry command-backed input values (GHSA-crhv-2887-p3xr). A fourth, GHSA-w9fm-7vcm-cvq4, is a security feature bypass in the Agent Host. Local-command syntax in an agent message could be treated as if the user typed it and run without the expected terminal approval. The advisory says the attacker must influence an agent's child prompt, typically through untrusted repository instructions, and the user must grant the related session tool operation. Each one needs some user action or a hostile MCP server, and the advisories list workarounds: connect only to trusted MCP servers, keep terminal approvals on, and avoid session-wide tool grants in repositories with untrusted instructions. Together they show that the MCP client and the agent loop in an IDE are now part of the attack surface. What to check: VS Code version on managed laptops (1.145.3 or later), the list of MCP servers configured in developer profiles, and whether policy blocks session-wide approvals. ## Earlier and later in the release log - [Anthropic patches Claude Desktop Cowork folder file-open flaw](/news/claude-desktop-cowork-folder-file-open-advisory/): 25 Sep 2026 - [Four more VS Code extensions reuse SaassyCode malware with a new delivery chain](/news/saassycode-vs-code-extensions-repackaged/): 24 Aug 2026 - [Bay emerges from stealth with $50M in total funding](/news/bay-emerges-from-stealth-with-50m/): 10 Nov 2026 - [Vanta signs agreement to acquire Pillar Security](/news/vanta-to-acquire-pillar-security/): 9 Nov 2026 Every dated item is on the [release log](/news/). Seen a release this log missed? [Send a tip](/corrections/). ## Sources - github.com/microsoft/vscode/security/advisories/GHSA-4x35-6qr8-ww48: https://github.com/microsoft/vscode/security/advisories/GHSA-4x35-6qr8-ww48 (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/news/vscode-1-145-3-mcp-and-agent-host-advisories/ --- # Bay emerges from stealth with $50M in total funding NEWS, 10 NOV 2026, read 12 Nov 2026 **The short answer.** Bay announced on 10 Nov 2026 that it has left stealth with $50 million in total funding: a $37 million Series A led by Battery Ventures and a $13 million seed. The company says it covers AI on employee devices, cloud and SaaS, and deploys without a persistent endpoint agent, kernel driver or network proxy. Bay's press release, dated 10 Nov 2026 from Tel Aviv, says the company came out of stealth less than a year after founding. It names customers including Micron, Peloton, ClickUp and Wix. The funding is a $37 million Series A led by Battery Ventures and a previously undisclosed $13 million seed led by YL Ventures and Striker, for $50 million in total. Bay says it brings AI governance and enforcement across employee devices, cloud infrastructure and SaaS applications in under 5 minutes, without an additional persistent endpoint agent, kernel driver or network proxy. It lists AI discovery and posture management, session-level observability and real-time enforcement. These are Bay's own claims from the release. The deployment model is the part to test. Agentless coverage of laptops trades some depth of local visibility for speed of rollout, while an installed sensor can see process-level activity. Which one fits depends on how much of your risk is in local coding agents versus SaaS copilots. What to check: in a pilot, which agents and MCP servers the product finds on a managed laptop, what it can block before a tool call runs, and what remains visible when a device is offline. ## Earlier and later in the release log - [Four more VS Code extensions reuse SaassyCode malware with a new delivery chain](/news/saassycode-vs-code-extensions-repackaged/): 24 Aug 2026 - [VS Code 1.145.3 fixes MCP client and agent host code execution flaws](/news/vscode-1-145-3-mcp-and-agent-host-advisories/): 10 Nov 2026 - [Vanta signs agreement to acquire Pillar Security](/news/vanta-to-acquire-pillar-security/): 9 Nov 2026 - [Knostic flags two npm MCP packages that collect data when launched](/news/two-malicious-npm-mcp-packages-hidden-collection/): 9 Nov 2026 Every dated item is on the [release log](/news/). Seen a release this log missed? [Send a tip](/corrections/). ## Sources - www.bay.io/blog/bay-emerges-from-stealth-with-50m-after-fortune-100-companies-used-its-platform-to-triple-autonomous-ai-agent-use-2: https://www.bay.io/blog/bay-emerges-from-stealth-with-50m-after-fortune-100-companies-used-its-platform-to-triple-autonomous-ai-agent-use-2 (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/news/bay-emerges-from-stealth-with-50m/ --- # Vanta signs agreement to acquire Pillar Security NEWS, 9 NOV 2026, read 12 Nov 2026 **The short answer.** Pillar Security announced on 9 Nov 2026 that Vanta signed a definitive agreement to acquire it. Pillar, founded in 2023, covers AI discovery, red teaming, runtime guardrails and agentic endpoint governance, and says its SAIL framework passed 50,000 downloads and maps 91 AI risks. Terms were not stated. Pillar Security's founders wrote that Vanta has signed a definitive agreement to acquire the company. The post says Pillar discovers, tests and secures AI applications and agents across their lifecycle and blocks unauthorized actions at runtime. It gives no price or closing date. Pillar says its research has found vulnerabilities rated CVSS 10, including an n8n sandbox escape and a Gemini CLI flaw that let one GitHub issue compromise the tool's supply chain. It also reports that its SAIL framework has passed 50,000 downloads and maps 91 AI risks across seven lifecycle phases. Vanta describes itself as an Agentic Trust Platform serving more than 20,000 companies. The stated plan is to connect Pillar's testing and runtime enforcement to Vanta's policy and evidence graph, so a written AI policy can drive controls and produce proof. For buyers, the practical question is product continuity: whether Pillar's agentic endpoint module stays a standalone product, how it is licensed after closing, and where roadmap priority sits between governance evidence and endpoint enforcement. Ask for those in writing before renewal. ## Earlier and later in the release log - [VS Code 1.145.3 fixes MCP client and agent host code execution flaws](/news/vscode-1-145-3-mcp-and-agent-host-advisories/): 10 Nov 2026 - [Bay emerges from stealth with $50M in total funding](/news/bay-emerges-from-stealth-with-50m/): 10 Nov 2026 - [Knostic flags two npm MCP packages that collect data when launched](/news/two-malicious-npm-mcp-packages-hidden-collection/): 9 Nov 2026 - [MCP Inspector fixed for remote code execution through a hostile server's OAuth endpoint](/news/mcp-inspector-rce-over-oauth-fixed-2-12-0/): 5 Nov 2026 Every dated item is on the [release log](/news/). Seen a release this log missed? [Send a tip](/corrections/). ## Sources - www.pillar.security/blog/vanta-to-acquire-pillar-security-building-trust-in-ai-together: https://www.pillar.security/blog/vanta-to-acquire-pillar-security-building-trust-in-ai-together (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/news/vanta-to-acquire-pillar-security/ --- # Knostic flags two npm MCP packages that collect data when launched NEWS, 9 NOV 2026, read 12 Nov 2026 **The short answer.** Knostic published a static review on 9 Nov 2026 of @ifasn/mmdd-doc-mcp and @ifasn/ms-doc-mcp, both version 2.1.1. Posing as Markdown toolkits, both start a hidden routine at launch. One tries to send process data to a hardcoded IP over HTTP; the other logs secret values. Knostic found no confirmed victims. Knostic's research note says two npm packages from the same account present themselves as Markdown toolkits for AI agents. Both start a collection routine whenever the MCP server launches, without any tool call. Knostic's AgentMesh scanner rated the first as Dangerous, and a manual review then covered the second. According to Knostic, mmdd-doc-mcp gathers process information, including command lines that may carry secrets, checks a cloud metadata service, and tries to send the first 1,500 characters of a report over plain HTTP to a hardcoded address. ms-doc-mcp records full values of environment variables with secret-like names in local output, and Knostic found no external server in its code. Knostic stresses this is static analysis with no confirmed delivery or victims. A local stdio MCP server runs as a process with the user's permissions, so it sees the same environment variables and process list as the developer. Installing one is closer to installing a binary than adding a plugin. What to check: which MCP packages are configured on developer machines, whether installs are limited to an approved list, and whether secrets are kept out of environment variables and command-line arguments that any local process can read. ## Earlier and later in the release log - [Bay emerges from stealth with $50M in total funding](/news/bay-emerges-from-stealth-with-50m/): 10 Nov 2026 - [Vanta signs agreement to acquire Pillar Security](/news/vanta-to-acquire-pillar-security/): 9 Nov 2026 - [MCP Inspector fixed for remote code execution through a hostile server's OAuth endpoint](/news/mcp-inspector-rce-over-oauth-fixed-2-12-0/): 5 Nov 2026 - [CrowdStrike adds AI Kill Switch and agent-level hooks to Falcon Guardian](/news/crowdstrike-falcon-guardian-ai-kill-switch/): 3 Nov 2026 Every dated item is on the [release log](/news/). Seen a release this log missed? [Send a tip](/corrections/). ## Sources - www.knostic.ai/blog/more-than-markdown-hidden-data-collection-in-two-mcp-packages: https://www.knostic.ai/blog/more-than-markdown-hidden-data-collection-in-two-mcp-packages (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/news/two-malicious-npm-mcp-packages-hidden-collection/ --- # MCP Inspector fixed for remote code execution through a hostile server's OAuth endpoint NEWS, 5 NOV 2026, read 12 Nov 2026 **The short answer.** Pillar Security reported on 5 Nov 2026 a High severity flaw in the official MCP Inspector on Windows and WSL. A remote MCP server could run code during OAuth login. The fix is in Inspector 2.12.0, tracked as GHSA-q978-78c7-phqw and CVE-2026-119025. The Inspector has about 768,000 weekly downloads. Pillar Security's write-up covers a remote code execution bug in the MCP Inspector, the debugging tool maintained by the Model Context Protocol project. A remote server answering with an OAuth challenge can name an authorization endpoint, and the client opened it with the open package. On Windows and WSL, open versions before 11.0.0 passed it to PowerShell unescaped. Pillar says the code ran before the browser opened and without user approval, since auto-open was on by default in an interactive terminal. The maintainers shipped 2.12.0 and published GHSA-q978-78c7-phqw, rated High at CVSS 3.1 8.8. Pillar says the same chain affected opencode, fixed in 1.18.33, and open 11.0.0 contains the root-cause fix. The notable point is scope. The Inspector exists to vet servers you do not trust, and the maintainers accepted that a hostile server is in scope. Any coding agent or IDE that implements the same OAuth flow should treat a remote MCP server as untrusted input. What to check: Inspector, opencode and open versions on Windows and WSL developer machines, and your policy on connecting agents to remote MCP servers that nobody on the security team has reviewed. ## Earlier and later in the release log - [Vanta signs agreement to acquire Pillar Security](/news/vanta-to-acquire-pillar-security/): 9 Nov 2026 - [Knostic flags two npm MCP packages that collect data when launched](/news/two-malicious-npm-mcp-packages-hidden-collection/): 9 Nov 2026 - [CrowdStrike adds AI Kill Switch and agent-level hooks to Falcon Guardian](/news/crowdstrike-falcon-guardian-ai-kill-switch/): 3 Nov 2026 - [CrowdStrike signs deal to acquire Fiddler AI for agent observability](/news/crowdstrike-to-acquire-fiddler-agent-observability/): 3 Nov 2026 Every dated item is on the [release log](/news/). Seen a release this log missed? [Send a tip](/corrections/). ## Sources - www.pillar.security/blog/inspecting-the-inspector-rce-in-mcps-official-debugging-tool: https://www.pillar.security/blog/inspecting-the-inspector-rce-in-mcps-official-debugging-tool (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/news/mcp-inspector-rce-over-oauth-fixed-2-12-0/ --- # CrowdStrike adds AI Kill Switch and agent-level hooks to Falcon Guardian NEWS, 3 NOV 2026, read 12 Nov 2026 **The short answer.** CrowdStrike announced at Fal.Con Europe on 3 Nov 2026 an AI Kill Switch for Falcon Guardian, plus agent-level hooks for Claude Code, OpenAI Codex, GitHub Copilot, Microsoft 365 Copilot, OpenCode and Cursor. Endpoint network inspection is generally available for Windows, with Mac in early access. CrowdStrike's 3 Nov 2026 press release says Falcon Guardian has been generally available since September 2026 and now adds an AI Kill Switch that terminates unauthorized agents, first at the gateway and soon on the endpoint, and blocks them from running again across the organization. The AI Gateway is now generally available and included with Guardian. Per CrowdStrike, agent-level hooks enforce policy inside Claude Code, OpenAI Codex, GitHub Copilot, Microsoft 365 Copilot, OpenCode and Cursor. The release also lists discovery of malicious AI skills and MCP servers, and a partnership with Anthropic to run the Falcon sensor inside the Claude Cowork virtual machine. Endpoint network inspection is generally available for Windows, with Mac in early access. CrowdStrike argues that hooks alone miss model responses, network-only tools go blind to local models, and a single sensor needs to connect all three. Those are the vendor's claims, and other vendors take different architectures. What to check: whether your EDR contract already includes the Guardian module, which agents on your fleet are on the hook list, and what the kill switch does today on the endpoint versus the gateway, since CrowdStrike says endpoint termination is still to come. ## Earlier and later in the release log - [Knostic flags two npm MCP packages that collect data when launched](/news/two-malicious-npm-mcp-packages-hidden-collection/): 9 Nov 2026 - [MCP Inspector fixed for remote code execution through a hostile server's OAuth endpoint](/news/mcp-inspector-rce-over-oauth-fixed-2-12-0/): 5 Nov 2026 - [CrowdStrike signs deal to acquire Fiddler AI for agent observability](/news/crowdstrike-to-acquire-fiddler-agent-observability/): 3 Nov 2026 - [Backdoored Claude Code extension build distributed through Cursor's marketplace](/news/poisoned-claude-code-extension-cursor-marketplace/): 30 Oct 2026 Every dated item is on the [release log](/news/). Seen a release this log missed? [Send a tip](/corrections/). ## Sources - www.crowdstrike.com/en-us/press-releases/crowdstrike-delivers-ai-security-safety-for-the-enterprise/: https://www.crowdstrike.com/en-us/press-releases/crowdstrike-delivers-ai-security-safety-for-the-enterprise/ (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/news/crowdstrike-falcon-guardian-ai-kill-switch/ --- # CrowdStrike signs deal to acquire Fiddler AI for agent observability NEWS, 3 NOV 2026, read 12 Nov 2026 **The short answer.** CrowdStrike said on 3 Nov 2026 it signed a definitive agreement to acquire Fiddler AI. Fiddler will be sold in the Falcon platform as Falcon Observe, generally available in Q4 FY27, adding pre-deployment evaluation, agent traceability and explainability next to the runtime controls in Falcon Guardian. CrowdStrike announced a definitive agreement to acquire Fiddler AI, a vendor of AI agent observability and explainability. The company says Fiddler will be sold as part of the Falcon platform as Falcon Observe, generally available in Q4 FY27. The release does not state a price or closing date. CrowdStrike lists three capabilities: pre-deployment evaluation of agent accuracy, task completion and policy compliance; agent traceability across decisions and reasoning paths; and explainability. CEO George Kurtz frames it as a record that lets a company answer a regulator or board about what an agent did. For endpoint teams the useful distinction is between blocking and accounting. Guardian-style controls stop an action on the machine. Observability tools reconstruct why an agent chose it. Most programs need both, and evidence of the second is what auditors ask for after an incident. What to check: whether your current tooling keeps a per-session record of prompts, tool calls and file changes for coding agents, how long it retains that record, and whether it can be exported to your SIEM. ## Earlier and later in the release log - [MCP Inspector fixed for remote code execution through a hostile server's OAuth endpoint](/news/mcp-inspector-rce-over-oauth-fixed-2-12-0/): 5 Nov 2026 - [CrowdStrike adds AI Kill Switch and agent-level hooks to Falcon Guardian](/news/crowdstrike-falcon-guardian-ai-kill-switch/): 3 Nov 2026 - [Backdoored Claude Code extension build distributed through Cursor's marketplace](/news/poisoned-claude-code-extension-cursor-marketplace/): 30 Oct 2026 - [Island launches its enterprise browser as a governed surface for AI agents](/news/island-enterprise-browser-for-ai-agents/): 29 Oct 2026 Every dated item is on the [release log](/news/). Seen a release this log missed? [Send a tip](/corrections/). ## Sources - www.crowdstrike.com/en-us/press-releases/crowdstrike-to-acquire-fiddler-to-deliver-agent-observability/: https://www.crowdstrike.com/en-us/press-releases/crowdstrike-to-acquire-fiddler-to-deliver-agent-observability/ (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/news/crowdstrike-to-acquire-fiddler-agent-observability/ --- # Backdoored Claude Code extension build distributed through Cursor's marketplace NEWS, 30 OCT 2026, read 12 Nov 2026 **The short answer.** Noma Security reported on 30 Oct 2026 that anthropic.claude-code version 2.1.313 was tampered with and distributed through Cursor's extension marketplace. Noma says it harvested environment metadata, planted an SSH key, opened reverse SSH tunnels and set up persistence on macOS, Linux and Windows. Version 2.1.314 or later is clean. Noma Security's research post describes a backdoored build of the anthropic.claude-code extension, version 2.1.313, that reached developer workstations through Cursor's extension marketplace before it was pulled. Noma says the attackers prepended a 66-line obfuscated block to the extension's entry-point bundle, so it ran when the IDE activated the extension, with no prompt or permission grant needed. According to Noma, the payload collected host details and up to 40 environment variable names matching keywords such as KEY, TOKEN and SECRET, sent them over plain HTTP, started the SSH service, appended an attacker key to authorized_keys, opened a reverse tunnel and installed persistence through a LaunchAgent, cron job or scheduled task. Noma also says Cursor published remediation scripts for macOS and Windows. This is ordinary supply chain code execution, not a prompt attack. It matters because an AI coding extension runs inside the IDE with the developer's credentials, so one bad update reaches tokens, keys and repositories. What to check: extension versions across the fleet (Noma gives commands for VS Code and Cursor), unexpected entries in authorized_keys, new LaunchAgents, cron entries or scheduled tasks named like system monitors, and outbound SSH from developer laptops. Updating alone does not remove the persistence. ## Earlier and later in the release log - [CrowdStrike adds AI Kill Switch and agent-level hooks to Falcon Guardian](/news/crowdstrike-falcon-guardian-ai-kill-switch/): 3 Nov 2026 - [CrowdStrike signs deal to acquire Fiddler AI for agent observability](/news/crowdstrike-to-acquire-fiddler-agent-observability/): 3 Nov 2026 - [Island launches its enterprise browser as a governed surface for AI agents](/news/island-enterprise-browser-for-ai-agents/): 29 Oct 2026 - [Claude Code plugin install could bypass a pinned commit SHA](/news/claude-code-plugin-sha-pin-bypass-advisory/): 9 Oct 2026 Every dated item is on the [release log](/news/). Seen a release this log missed? [Send a tip](/corrections/). ## Sources - noma.security/blog/supply-chain-attack-poisoned-claude-code-extension-installs-hidden-backdoors: https://noma.security/blog/supply-chain-attack-poisoned-claude-code-extension-installs-hidden-backdoors (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/news/poisoned-claude-code-extension-cursor-marketplace/ --- # Island launches its enterprise browser as a governed surface for AI agents NEWS, 29 OCT 2026, read 12 Nov 2026 **The short answer.** Island announced on 29 Oct 2026 an enterprise browser for AI agents. It sets per-agent, per-application policy for browser MCP tools, ties agent actions to the user's identity through IdP federation, and labels each audit log event as user or agent. It runs on the device or in the cloud, headless or rendered. Island's post argues that the common setup, a consumer browser plus an AI provider's extension exposing browser tools over MCP, has no policy layer, cannot tell agent actions from user actions and can expose credentials. Island's answer is to make the browser itself the control point for agents. Per Island, admins can define which agents may use the browser, which pages they may touch, which DOM operations are allowed and whether an agent can read or also write. Agents inherit the user's identity, group and device posture, and policy can target the user, the user plus agents, or agents only. Island says every audit log event is attributed to the human or to an agent acting for them, which it ties to SOC 2, ISO 27001, HIPAA and PCI DSS evidence. It says the browser runs on the device or in the cloud, headless or fully rendered. This covers browser agents only. Terminal agents, IDE agents and local MCP servers sit outside a browser, so a security lead still needs a way to see and control those. What to check: which browser extensions with agent capability are installed on managed devices, and whether your current browser policy can distinguish an agent click from a human one. ## Earlier and later in the release log - [CrowdStrike signs deal to acquire Fiddler AI for agent observability](/news/crowdstrike-to-acquire-fiddler-agent-observability/): 3 Nov 2026 - [Backdoored Claude Code extension build distributed through Cursor's marketplace](/news/poisoned-claude-code-extension-cursor-marketplace/): 30 Oct 2026 - [Claude Code plugin install could bypass a pinned commit SHA](/news/claude-code-plugin-sha-pin-bypass-advisory/): 9 Oct 2026 - [Anthropic patches Claude Desktop Cowork folder file-open flaw](/news/claude-desktop-cowork-folder-file-open-advisory/): 25 Sep 2026 Every dated item is on the [release log](/news/). Seen a release this log missed? [Send a tip](/corrections/). ## Sources - www.island.io/blog/island-the-enterprise-browser-for-ai-agents: https://www.island.io/blog/island-the-enterprise-browser-for-ai-agents (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/news/island-enterprise-browser-for-ai-agents/ --- # Claude Code plugin install could bypass a pinned commit SHA NEWS, 9 OCT 2026, read 12 Nov 2026 **The short answer.** Anthropic disclosed on 9 Oct 2026 that the plugin install flow in Claude Code and Claude Desktop checked out a marketplace's pinned commit without confirming HEAD matched it. A branch named like the 40-character SHA could resolve to attacker code. Rated High, CVSS 7.7, with a fix shipped through auto-update. Anthropic's advisory GHSA-pq7j-f95f-qfcv covers the plugin install flow in Claude Code and Claude Desktop. The flow ran git checkout on a marketplace's pinned commit SHA but did not verify that the resulting HEAD matched. Someone controlling a plugin's upstream repository could create a branch with the same name as the 40-character SHA, and the checkout would land on the attacker's branch tip. Anthropic rates the issue High at 7.7 on CVSS v4. It says exploitation needs a user to install a malicious plugin, pin it to a safe-looking hash, and a git host that allows SHA-shaped branch names. Anthropic credits Or Nevo Michrowski of Air Security. Standard auto-update has delivered the fix; manual updates need the latest version. The lesson for endpoint teams is that a pinned hash is only a control if the installer verifies it. Plugins run with the permissions of the agent that loads them, so a swapped plugin is code execution inside a tool that already holds repository access and often cloud credentials. What to check: that developer machines are on the current Claude Code release, which plugin marketplaces are allowed in managed settings, and whether you have an inventory of installed plugins and the commit each one resolves to. ## Earlier and later in the release log - [Backdoored Claude Code extension build distributed through Cursor's marketplace](/news/poisoned-claude-code-extension-cursor-marketplace/): 30 Oct 2026 - [Island launches its enterprise browser as a governed surface for AI agents](/news/island-enterprise-browser-for-ai-agents/): 29 Oct 2026 - [Anthropic patches Claude Desktop Cowork folder file-open flaw](/news/claude-desktop-cowork-folder-file-open-advisory/): 25 Sep 2026 - [Four more VS Code extensions reuse SaassyCode malware with a new delivery chain](/news/saassycode-vs-code-extensions-repackaged/): 24 Aug 2026 Every dated item is on the [release log](/news/). Seen a release this log missed? [Send a tip](/corrections/). ## Sources - github.com/anthropics/claude-code/security/advisories/GHSA-pq7j-f95f-qfcv: https://github.com/anthropics/claude-code/security/advisories/GHSA-pq7j-f95f-qfcv (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/news/claude-code-plugin-sha-pin-bypass-advisory/ --- # Anthropic patches Claude Desktop Cowork folder file-open flaw NEWS, 25 SEP 2026, read 12 Nov 2026 **The short answer.** Anthropic published a High severity advisory (CVSS 8.5) on 25 Sep 2026 for Claude Desktop on macOS. A file placed in a Cowork shared folder by a prompt-injected agent could run commands when the user opened it. Claude Desktop 1.15962.0 extends the block list of executable file types. Anthropic's GitHub advisory GHSA-v234-4jrq-mgg6 describes a gap in the list of file types that Claude Desktop refuses to open from a Cowork session's shared folder. On macOS the list left out one file type that the operating system executes on open. Anthropic says a compromised or prompt-injected agent could drop such a file, and the user opening it from Claude Desktop would run commands on the Mac. The advisory also notes that Cowork VM images before Claude Desktop 1.11847.5 (released 9 June 2026) shipped a guest Linux kernel affected by CVE-2026-43284. Anthropic says the two issues combined could let code with higher privileges inside the VM trigger the file open without user interaction. The severity rating covers the file-handling flaw alone. Anthropic says it found the issue internally and that Vladimir Tokarev of Cyera Research also reported it. For a security lead, the point is that an agent's sandbox and the host are separate trust zones, and a shared folder is the bridge. Files an agent writes there are untrusted input. Standard auto-update has already delivered 1.15962.0, but the advisory says managed or manual update paths need action. What to check: the Claude Desktop version on every managed Mac and Windows machine, whether your update channel pins old builds, and whether endpoint tooling can see file opens that originate from a Cowork folder. ## Earlier and later in the release log - [Island launches its enterprise browser as a governed surface for AI agents](/news/island-enterprise-browser-for-ai-agents/): 29 Oct 2026 - [Claude Code plugin install could bypass a pinned commit SHA](/news/claude-code-plugin-sha-pin-bypass-advisory/): 9 Oct 2026 - [Four more VS Code extensions reuse SaassyCode malware with a new delivery chain](/news/saassycode-vs-code-extensions-repackaged/): 24 Aug 2026 - [VS Code 1.145.3 fixes MCP client and agent host code execution flaws](/news/vscode-1-145-3-mcp-and-agent-host-advisories/): 10 Nov 2026 Every dated item is on the [release log](/news/). Seen a release this log missed? [Send a tip](/corrections/). ## Sources - github.com/anthropics/claude-code/security/advisories/GHSA-v234-4jrq-mgg6: https://github.com/anthropics/claude-code/security/advisories/GHSA-v234-4jrq-mgg6 (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/news/claude-desktop-cowork-folder-file-open-advisory/ --- # Four more VS Code extensions reuse SaassyCode malware with a new delivery chain NEWS, 24 AUG 2026, read 12 Nov 2026 **The short answer.** Knostic reported on 24 Aug 2026 four VS Code Marketplace extensions that reuse code from the SaassyCode family: ManageRBLX123, FocusDeck, Trello Deck and Swimlane. Three source derivatives had 2,254 observed installs. Their loader replaces earlier indicators such as a cscript.exe child process and a nice.js file in the temp directory. Knostic says it found four VS Code extensions that reuse malicious code from the SaassyCode family, which its earlier research counted at 24 extensions. One, ManageRBLX123, is an almost exact repackaging that differs from an original by five lines of metadata. The other three, FocusDeck, Trello Deck and Swimlane, are built on TrelloBlox 5.7.0 source. Knostic says the three derivatives removed the original dropper and now spawn a detached Node process using the editor's own binary, which runs an install script in a bundled dependency that downloads and executes a payload. That removes the cscript.exe child process and the nice.js file that earlier detections looked for. The observed install counts were 1,015, 55 and 1,184 as of 19 to 20 Aug 2026, and Knostic notes these are not confirmed infections. For a security lead, the lesson is that detections keyed on one campaign's artifacts age quickly, and marketplace listings are cheap to recreate. Extensions run with the developer's rights inside the same editor that hosts coding agents. What to check: an allowlist of approved extensions, alerts on the editor binary spawning a detached child that makes network calls, and inventory of extension publishers across the fleet. ## Earlier and later in the release log - [Claude Code plugin install could bypass a pinned commit SHA](/news/claude-code-plugin-sha-pin-bypass-advisory/): 9 Oct 2026 - [Anthropic patches Claude Desktop Cowork folder file-open flaw](/news/claude-desktop-cowork-folder-file-open-advisory/): 25 Sep 2026 - [VS Code 1.145.3 fixes MCP client and agent host code execution flaws](/news/vscode-1-145-3-mcp-and-agent-host-advisories/): 10 Nov 2026 - [Bay emerges from stealth with $50M in total funding](/news/bay-emerges-from-stealth-with-50m/): 10 Nov 2026 Every dated item is on the [release log](/news/). Seen a release this log missed? [Send a tip](/corrections/). ## Sources - www.knostic.ai/blog/saassycode-repackaged-four-new-vs-code-extensions-and-a-new-delivery-chain: https://www.knostic.ai/blog/saassycode-repackaged-four-new-vs-code-extensions-and-a-new-delivery-chain (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/news/saassycode-vs-code-extensions-repackaged/ --- # 12 AI agent security releases and advisories, each from its primary source NEWS, read 12 Nov 2026 **The short answer.** Dated AI agent security releases, advisories and product announcements from primary sources, from 24 Aug 2026 to 10 Nov 2026. Each item links to the original vendor or project page, carries the date it was read, and is written in plain sentences with names and numbers. ## Release log Newest first: 12 news items from primary sources, the dates the evidence was read and the dates the weights and scores were fixed. Each line links to the page that uses it. - 13 Nov 2026: [Blind scores published for 16 tools](/ranking/). Leader on the published weights: Microsoft at 75.0. - 13 Nov 2026: [Second evidence pass, one plan for every tool: 169 more facts from 103 source pages](/method/). Every fact is in the evidence dataset with its source URL. - 12 Nov 2026: [First evidence read: 792 facts from 328 source pages across 16 tools](/method/). Every fact is in the evidence dataset with its source URL. - 12 Nov 2026: [Weights fixed and committed before any vendor was scored](/method/). Sees 25, Stops 25, Deployment 15, Beyond 10, Policy 10, Transparency 10, Proof 5. - 10 Nov 2026: [Bay emerges from stealth with $50M in total funding](/news/bay-emerges-from-stealth-with-50m/). Primary source: bay.io. - 10 Nov 2026: [VS Code 1.145.3 fixes MCP client and agent host code execution flaws](/news/vscode-1-145-3-mcp-and-agent-host-advisories/). Primary source: github.com. - 9 Nov 2026: [Knostic flags two npm MCP packages that collect data when launched](/news/two-malicious-npm-mcp-packages-hidden-collection/). Primary source: knostic.ai. - 9 Nov 2026: [Vanta signs agreement to acquire Pillar Security](/news/vanta-to-acquire-pillar-security/). Primary source: pillar.security. - 5 Nov 2026: [MCP Inspector fixed for remote code execution through a hostile server's OAuth endpoint](/news/mcp-inspector-rce-over-oauth-fixed-2-12-0/). Primary source: pillar.security. - 3 Nov 2026: [CrowdStrike adds AI Kill Switch and agent-level hooks to Falcon Guardian](/news/crowdstrike-falcon-guardian-ai-kill-switch/). Primary source: crowdstrike.com. - 3 Nov 2026: [CrowdStrike signs deal to acquire Fiddler AI for agent observability](/news/crowdstrike-to-acquire-fiddler-agent-observability/). Primary source: crowdstrike.com. - 30 Oct 2026: [Backdoored Claude Code extension build distributed through Cursor's marketplace](/news/poisoned-claude-code-extension-cursor-marketplace/). Primary source: noma.security. - 29 Oct 2026: [Island launches its enterprise browser as a governed surface for AI agents](/news/island-enterprise-browser-for-ai-agents/). Primary source: island.io. - 9 Oct 2026: [Claude Code plugin install could bypass a pinned commit SHA](/news/claude-code-plugin-sha-pin-bypass-advisory/). Primary source: github.com. - 25 Sep 2026: [Anthropic patches Claude Desktop Cowork folder file-open flaw](/news/claude-desktop-cowork-folder-file-open-advisory/). Primary source: github.com. - 24 Aug 2026: [Four more VS Code extensions reuse SaassyCode malware with a new delivery chain](/news/saassycode-vs-code-extensions-repackaged/). Primary source: knostic.ai. Know of a release or an advisory this log is missing? [Send a tip with its source](/corrections/). ## Sources - Method: https://aiendpointsecurityranked.com/method/ (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/news/ --- # 30 AI endpoint security questions, answered with names and numbers FAQ, read 12 Nov 2026 **The short answer.** Thirty questions about AI endpoint security answered with names, numbers and dates: what the category covers, how coding agents are secured on laptops, what agentless means, how the index is scored and what a vendor claim is worth until a second source confirms it. **What does AI endpoint security cover?** It covers AI agents, assistants, MCP servers and browser extensions that run on employee laptops, in browsers and in developer tools. The job has two parts: show which of them exist and what they do, then stop the ones that act outside policy. This site maps six surfaces: desktop AI apps, IDE assistants, command-line agents, MCP servers, browser extensions and SaaS copilots. **How is AI endpoint security different from EDR?** EDR records and responds to process, file and network behavior on a device. It generally cannot tell a shell command run by a coding agent from one typed by a developer, and it does not read prompts or tool calls. AI endpoint security adds agent-level context: which tool, which session, which tool call. Some products build on EDR telemetry. **How is it different from DLP?** DLP inspects content moving to a destination and was built around user actions such as paste, upload and copy. Microsoft's endpoint DLP, for example, has rules for paste to supported browsers and upload to restricted domains. Agents add reads and tool calls with no user action, often from a terminal. AI endpoint security aims at those actions, not only at the destination. **How is it different from CASB?** A CASB watches SaaS use through APIs, proxies or logs. It sees cloud apps and OAuth-connected services, not what runs on the laptop. A local MCP server or a command-line agent using a personal API key never touches the SaaS API. CASB data is useful for the SaaS copilot surface and complements endpoint visibility. **Which agent surfaces matter most?** It depends on who works at your company. For engineering teams, command-line agents, IDE assistants and local MCP servers carry the highest-impact actions: shell commands, credential access and package installs. For knowledge workers, browser extensions, desktop AI apps and SaaS copilots hold more of the data-paste and oversharing risk. Most companies need to cover all six eventually. **What actions should a tool be able to see?** Six matter: file read, shell command, network call, credential access, data paste and package install. A product should say, for each surface, which of these it sees, which it can stop, and which component on the device does so. A product that lists surfaces without listing actions has not told you what it covers. **What does blocking before completion mean?** It means the product decides whether an action may proceed before the action finishes: the command does not run, the file is not read, the paste does not land. Alerting after the fact is detection. Both are useful, but they are different. Ask a vendor to demonstrate a block on your own machine, and to show what the user sees. **Why is the difference between detection and prevention important here?** An agent can read a credential file or run a package install in seconds, and an alert that arrives an hour later cannot recall what was sent. OWASP names excessive functionality, permissions and autonomy as root causes of agent harm, which a pre-action check can limit. Detection still matters for inventory and investigation. **What does agentless mean for AI endpoint security?** It usually means no new sensor ships to the laptop. The data may come from an MDM profile, a browser extension, an OS API, existing EDR telemetry or a SaaS API connector. Each sees a different slice. Browser extensions miss terminal agents, and API connectors miss processes. Ask which component produces the data for each surface. **Can MDM alone manage AI tools on laptops?** MDM can push settings to tools that read managed policy. Anthropic documents that Claude Code accepts endpoint-managed settings through macOS managed preferences, Windows registry or a managed file. Intune can apply custom macOS profiles. But MDM is a configuration channel, not a sensor: it does not show what the tools do or find tools you did not know about. **Where do local MCP servers live on a laptop?** They are defined in each client's config file. Claude Desktop uses claude_desktop_config.json (under Application Support on macOS, APPDATA on Windows, .config on Linux). Cursor uses .cursor/mcp.json in a project and ~/.cursor/mcp.json globally. Claude Code uses .mcp.json in a project and ~/.claude.json for user and local scope. **Why are local MCP servers a risk?** A stdio MCP server is a program the client launches with the user's privileges. The MCP security best practices page lists attacks including a malicious startup command in a client configuration and a malicious payload inside the server, and asks for exact-command consent and sandboxing with restricted file system and network access. **How do I control MCP servers on laptops?** Control is per client. Claude Code reads a managed-mcp.json from a fixed system path for exclusive control, plus allow and deny lists. VS Code has a ChatMCP access policy and an allowed-servers list. For clients without an equivalent policy, none found on 12 Nov 2026 in the docs read, so monitor the config files and block launch commands. **Does a project folder's MCP config add risk?** Yes. Cursor and Claude Code both read an MCP config from inside the project (.cursor/mcp.json, .mcp.json), and project files can be committed to source control. A cloned repository can propose a server to a developer's machine. Claude Code has project approval and workspace trust steps for this, and developers need to read those prompts. **What is the difference between stdio and remote MCP servers?** With stdio, the client launches the server as a subprocess on the same machine and talks to it over standard input and output. A remote server is reached over Streamable HTTP. The MCP specification defines both transports. Stdio servers are the endpoint concern, because they run locally with the user's rights. **What risks do coding agents bring to developer machines?** Coding agents run shell commands, read the project and the home directory, install packages and call network endpoints, all as the developer. Developer laptops often hold source code, signing keys and cloud credentials. The main risks are excessive permissions, unreviewed MCP servers and credential exposure in project files. Managed settings and sandboxing reduce them. **Can I enforce settings on Claude Code across a fleet?** Anthropic says Claude Code supports endpoint-managed settings deployed through native OS policies (macOS managed preferences, Windows registry) or managed settings files, and server-managed settings fetched from Anthropic's servers for organizations without MDM. Endpoint-managed settings are the stronger route where you run MDM. **What does VS Code offer for enterprise control of agents?** Microsoft publishes VS Code enterprise policies including ChatMCP (access to installed MCP servers), ChatAllowedMcpServers (an allowlist, with deny rules taking precedence), and policies for agent hooks and plugins. These are deployed through your device management tooling. They govern the editor's features, not other tools on the machine. **What are browser agents, and why do they need separate controls?** Browser agents are AI features that read pages and act on them: click, fill forms and move across tabs. Examples include Perplexity's Comet. They act inside the user's signed-in sessions. Anthropic's November 2025 post says no browser agent is immune to prompt injection. Controls are extension policy, site scoping and confirmation steps. **How do AI browser extensions get their access?** Through the permissions in their manifest. Google's documentation explains that permission warnings describe API capabilities and that a broad host permission lets an extension act on every site. An AI extension that reads pages usually requests that breadth. Review permissions, publisher and update history before allowing it, and restrict by policy. **Can I control which extensions run in Chrome and Edge?** Yes. Both support the ExtensionSettings policy, which can control update URLs and blocked permissions, and sits alongside install allowlists, blocklists and force-install lists. Microsoft documents it for Edge. The strongest setup is default deny with a reviewed allowlist by extension ID. Also manage Incognito so extensions are not bypassed. **What does AI DLP on the endpoint do?** It applies data loss rules to AI-related actions: pasting text into an AI site, uploading a file, or copying from protected files. Purview endpoint DLP, for example, evaluates pasted content in supported browsers and can warn, block or audit. It covers people at keyboards well. Agent reads and API calls from terminals are the thinner area. **Does Microsoft Purview block pasting into ChatGPT?** Microsoft's documentation says Purview DLP policies can warn or block users from sharing sensitive information with third-party generative AI sites accessed through a browser, with pasting credit card numbers into ChatGPT as its example. For Chrome, the Purview extension must be installed. Check the current platform matrix, since some entries are marked preview. **How do I detect shadow AI on endpoints?** Combine signals: process and binary inventory, browser extension inventory, network and DNS records, agent and MCP config files, and OAuth grants. Each sees a slice. DNS shows a domain but not a prompt. OAuth grants miss personal accounts. A tool installed on purpose in a test run shows which signals work. **Which OAuth grants should I review for AI apps?** Start with apps holding write or full-mailbox scopes, then any app consented by several users in a short period. Microsoft documents how to review the permissions granted to applications in a Microsoft Entra tenant, including apps added through user or admin consent. This catches apps on unmanaged devices but not personal accounts or API keys. **What should I ask an AI endpoint security vendor?** Ask which surfaces and which actions it covers, which component on the device produces the data, whether it can block before an action completes, what a user can do to disable it, and how it handles a tool launched from a terminal. Ask for a trial where you run one tool per surface and compare its log with what you did. **How do I test a vendor claim?** Run one tool per surface on a test machine: a desktop app, an IDE assistant, a command-line agent, a local MCP server, a browser extension and a SaaS copilot sign-in. Make each do something harmless and observable. Compare the product's record with your actions, then ask to see a block. Score what you saw, not the data sheet. **How does this site score vendors?** Seven criteria with fixed weights set before any evidence was scored: sees 25, stops 25, deployment 15, beyond the endpoint 10, policy and governance 10, transparency 10 and proof 5. Each criterion is scored from sourced evidence. The weights are the same for every vendor. **What happens to a vendor claim that nobody else confirms?** A vendor's own claim is capped at 7 of 10 on a criterion until a second source confirms it, such as a write-up from another publisher or a public artefact like documentation, code or a release record. The claim is written as the vendor's. Once confirmed, it can score higher. This keeps marketing copy from scoring the same as verified behavior. **What score does a missing feature get?** A feature with no evidence found scores 0 on that criterion. The page records it as none found on a date, for example none found on 12 Nov 2026, rather than as a statement that the vendor lacks it. If the vendor later publishes evidence, the score is revisited and the date updates. ## Where the answers come from - [The 12 guides](/guides/) - [How the index scores](/method/) - [The glossary](/glossary/) - [The ranked index](/ranking/) ## Sources - Anthropic: Claude Code docs, security: https://code.claude.com/docs/en/security (read 12 Nov 2026) - Anthropic: Claude Code docs, permissions: https://code.claude.com/docs/en/permissions (read 12 Nov 2026) - Cursor: security: https://cursor.com/security (read 12 Nov 2026) - Cursor: docs, agent security: https://cursor.com/docs/agent/security (read 12 Nov 2026) - Microsoft: VS Code docs, Copilot security: https://code.visualstudio.com/docs/copilot/security (read 12 Nov 2026) - Microsoft: VS Code docs, agent tools: https://code.visualstudio.com/docs/copilot/agents/agent-tools (read 12 Nov 2026) - OpenAI: Codex docs, agent approvals and security: https://developers.openai.com/codex/agent-approvals-security (read 12 Nov 2026) - OpenAI: Codex docs, configuration reference: https://developers.openai.com/codex/config-reference (read 12 Nov 2026) - Google: github.com/google-gemini/gemini-cli docs/cli/sandbox.md: https://raw.githubusercontent.com/google-gemini/gemini-cli/main/docs/cli/sandbox.md (read 12 Nov 2026) - Google: github.com/google-gemini/gemini-cli docs/cli/trusted-folders.md: https://raw.githubusercontent.com/google-gemini/gemini-cli/main/docs/cli/trusted-folders.md (read 12 Nov 2026) - Anthropic: Claude Code docs, sandboxing: https://code.claude.com/docs/en/sandboxing (read 12 Nov 2026) - Microsoft Learn: Intune custom settings for macOS: https://learn.microsoft.com/en-us/mem/intune/configuration/custom-settings-macos (read 12 Nov 2026) - Anthropic: Claude Code server-managed settings: https://code.claude.com/docs/en/server-managed-settings (read 12 Nov 2026) - modelcontextprotocol.io: security best practices: https://modelcontextprotocol.io/docs/tutorials/security/security_best_practices (read 12 Nov 2026) - modelcontextprotocol.io: transports specification 2025-11-25: https://modelcontextprotocol.io/specification/2025-11-25/basic/transports (read 12 Nov 2026) - Perplexity: Comet: https://www.perplexity.ai/comet (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/faq/ --- # Four datasets, CSV and JSON, free to reuse with a citation DATASETS, read 13 Nov 2026 **The short answer.** AI Endpoint Security Ranked publishes four datasets under CC BY 4.0, read 12 Nov 2026: the Agent Surface Map and the Action Control Matrix (16 tools, six cells each), the scores (16 tools, seven criteria, with weights), and the evidence (961 facts, each with its source URL and read date). | Dataset | Rows | CSV | JSON | | --- | --- | --- | --- | | The Agent Surface Map | 16 | agent-surface-map.csv | agent-surface-map.json | | The Action Control Matrix | 16 | action-control-matrix.csv | action-control-matrix.json | | AI Endpoint Security Ranked scores | 112 | aes-scores.csv | aes-scores.json | | Score changes after each evidence pass | 93 | score-changes.csv | score-changes.json | | AI Endpoint Security Ranked evidence | 961 | aes-evidence.csv | aes-evidence.json | The weights file used for every score is published as [weights.json](/datasets/weights.json), and the ranked views as [ranked-views.json](/datasets/ranked-views.json). ## Cite this AI Endpoint Security Ranked, Agent Surface Map, Action Control Matrix and scores, edition of 13 Nov 2026, aiendpointsecurityranked.com/datasets/, CC BY 4.0. Each JSON file carries the edition, the read date, the license and the cite line. ## The assets behind the datasets - [The Agent Surface Map](/agent-surface-map/) - [The Action Control Matrix](/action-control-matrix/) - [How a surface cell is read](/agent-surface-map/methodology/) - [How an action cell is read](/action-control-matrix/methodology/) - [How the scores are made](/method/) ## Sources - Agent Surface Map methodology: https://aiendpointsecurityranked.com/agent-surface-map/methodology/ (read 12 Nov 2026) - Scoring method: https://aiendpointsecurityranked.com/method/ (read 12 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/datasets/ --- # An editorial ranked index of AI endpoint security tools ABOUT AI ENDPOINT SECURITY RANKED, read 12 Nov 2026 **The short answer.** AI Endpoint Security Ranked publishes one ranked index of AI endpoint security tools, two original datasets (the Agent Surface Map and the Action Control Matrix), profiles, head-to-head comparisons, guides and a glossary. Every fact carries a source URL and the date it was read, and the scoring weights are public. The publication covers AI endpoint security: the tools that show a security team which AI agents, coding assistants, MCP servers and browser extensions run on its machines, and that stop harmful agent actions before they complete. It does not cover AI security platforms in general, agent identity, or application security. A tool enters the index when public sources show it is sold or documented as an AI endpoint security product. Each is profiled from its own documentation and from third-party coverage, scored on seven criteria, and placed on the Agent Surface Map and the Action Control Matrix. The method is on the [Method page](/method/), the evidence rules are in the [editorial policy](/editorial-policy/), and every dataset can be downloaded from the [datasets page](/datasets/). Corrections are handled as described on the [corrections page](/corrections/). ## Related pages - [Editorial policy](/editorial-policy/) - [Corrections and tips](/corrections/) - [Contact the editor](/contact/) - [Privacy](/privacy/) - [Terms of use](/terms/) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/about/ --- # Every fact has a source, every claim is labelled, every weight is public EDITORIAL POLICY: SOURCES, LABELS AND PUBLIC WEIGHTS, read 12 Nov 2026 **The short answer.** Every page follows the same rules: a fact is published only with a source URL and the date it was read, a vendor's claim is labelled as the vendor's, a score is capped at 7 of 10 until a second source confirms it, and the weights are fixed before any score and never changed. A fact enters the data only with a quote, a source URL and a read date. A vendor's statement about its own product is written as the vendor's statement. A claim that only the vendor makes scores at most 7 of 10 on the criterion it supports; a second source, either a third-party write-up or something observed directly in public documentation or code, is needed for 8 to 10. Where nothing public names a capability, the page reads none found on the read date. That is a statement about the sources read, not about the product. Each competitor's strength is stated on its own profile in one sentence, and no fact about a competitor is left out to change a ranking. A score changes only when a sourced fact changes, and the change is logged with the fact. Security content follows one rule: risks, detection, policy and fixes, each sourced. The site publishes no exploit code, no prompt-injection payloads and no steps to bypass a control. Prices, claims and documentation older than 60 days are re-read, and the read date on the page moves only when the page was re-read. ## Related pages - [About the index](/about/) - [Corrections and tips](/corrections/) - [Contact the editor](/contact/) - [Privacy](/privacy/) - [Terms of use](/terms/) ## Sources - Method: https://aiendpointsecurityranked.com/method/ (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/editorial-policy/ --- # Send a factual correction with a source and it is checked within a week CORRECTIONS, read 13 Nov 2026 **The short answer.** Send a factual correction to editor@aiendpointsecurityranked.com with the page address and a source URL. Corrections are checked against the source within seven days, applied with a new read date where the source confirms them, and logged below. Scores change only when a sourced fact changes. Email: [editor@aiendpointsecurityranked.com](mailto:editor@aiendpointsecurityranked.com) A correction needs three things: the address of the page, the sentence or cell that is wrong, and a public source that shows what is right. The source can be a vendor page, a release note, a filing or a third-party write-up. When the source confirms the correction, the fact in the evidence file is changed with the new source and read date, the pages that cite it are rebuilt, and any score that rests on the fact is re-read under the same rubric. A score is never changed without a changed fact. If the source does not confirm the correction, the reply says what was read and why the page stands. Corrections log: none received as of 13 Nov 2026. ## Changelog 13 Nov 2026: a further evidence pass ran one search plan for all 16 tools alike (the same queries, source types and budget per tool), looking for second sources for scores at the 7-of-10 cap and for lower scores resting on vendor claims alone, items published since the first read and evidence against each current score. It added 134 facts from 78 new source pages and 35 records of a second source looked for and none found. The 87 cells those facts touched were re-scored blind and checked by a second blind reviewer; 20 scores moved, 15 up and 5 down. Weights and criteria did not change. Microsoft stays first at 75.0; Bay falls 0.5 to 71.5 (its Proof score fell from 9 to 8) and moves from 3rd to 2nd only because Harmonic Security fell further, from 72.5 to 71.5 (its Beyond score fell from 8 to 7); the two tie, both score 8 on Sees, the heaviest criterion, and name order lists Bay 2nd; Aim Security rises from 10th to 4th and Koi from 16th to 12th, and SentinelOne now leads WitnessAI in their head-to-head, 62.0 to 61.0 on the site weights and 63.5 to 61.5 for the staff rollout. Every change, with the facts that moved it, is on the [method page](/method/) and in the [score change log](/datasets/score-changes.csv). 13 Nov 2026: profile text corrected where the new evidence contradicted it: SentinelOne (Prompt Security) now has named customers on its own site, Koi publishes its operating systems, and the weakness lines of Aim Security, Bay, CrowdStrike, Harmonic Security and Noma Security carry the new limits found. 12 Nov 2026: first edition, 16 tools scored blind on seven criteria with weights fixed before the first score; an equal-depth evidence pass and a blind re-score the same day, and a re-scope of Palo Alto Networks. ## Send a tip A tip is a release, an advisory or a vendor change the [release log](/news/) has not picked up. Send the primary source URL to editor@aiendpointsecurityranked.com; a tip that a primary source confirms is added with its date, and any score it touches is re-read under the same rubric. ## Related pages - [About the index](/about/) - [Editorial policy](/editorial-policy/) - [Contact the editor](/contact/) - [Privacy](/privacy/) - [Terms of use](/terms/) ## Sources - Method: https://aiendpointsecurityranked.com/method/ (read 13 Nov 2026) Last reviewed 13 Nov 2026. Canonical: https://aiendpointsecurityranked.com/corrections/ --- # Write to the editor at editor@aiendpointsecurityranked.com CONTACT THE EDITOR, read 12 Nov 2026 **The short answer.** The editor can be reached at editor@aiendpointsecurityranked.com for factual corrections, questions about a score, requests to cite or reuse the datasets, and press enquiries. Corrections that include the page address and a public source URL are answered first, usually within seven days. Email: [editor@aiendpointsecurityranked.com](mailto:editor@aiendpointsecurityranked.com) Use one message per topic and include the page address. For a correction, add the source URL that shows the right fact; the [corrections page](/corrections/) explains what happens next. To reuse the Agent Surface Map, the Action Control Matrix, the scores or the evidence, the license is CC BY 4.0: cite the edition and link to the [datasets page](/datasets/). The site has no accounts, no forms and no newsletter, so there is nothing to sign up for. ## Related pages - [About the index](/about/) - [Editorial policy](/editorial-policy/) - [Corrections and tips](/corrections/) - [Privacy](/privacy/) - [Terms of use](/terms/) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/contact/ --- # No cookies, no analytics beacon, and server logs kept for 90 days PRIVACY: NO COOKIES AND NO ANALYTICS BEACON, read 12 Nov 2026 **The short answer.** AI Endpoint Security Ranked sets no cookies, runs no analytics script and loads no third-party fonts, scripts or trackers. The host, Amazon CloudFront, writes standard access logs with the time, address requested, IP address, user agent and referrer. Logs are kept for 90 days and used only for traffic counts and abuse checks. Pages are static files. Fonts, styles and images are served from the same address as the page. Nothing on the page sends data to another site. Access logs are written by the host for every request. They are read in aggregate to count visits and spot abuse, are not combined with any other data, and expire after 90 days. If you email the editor, the message and your address are kept only as long as needed to answer and to log any correction. Questions about this page go to editor@aiendpointsecurityranked.com. ## Related pages - [About the index](/about/) - [Editorial policy](/editorial-policy/) - [Corrections and tips](/corrections/) - [Contact the editor](/contact/) - [Terms of use](/terms/) ## Sources - Method: https://aiendpointsecurityranked.com/method/ (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/privacy/ --- # Read freely, reuse the datasets with a citation TERMS OF USE: READ FREELY, REUSE WITH A CITATION, read 12 Nov 2026 **The short answer.** The pages may be read, linked and quoted with attribution. The datasets (the Agent Surface Map, the Action Control Matrix, the scores and the evidence) are published under CC BY 4.0, which allows reuse with a citation of the edition. Facts are as read on the date shown, so check the source before you act. Cite the datasets as shown on the [datasets page](/datasets/). Quote pages with a link back to the page quoted. Product names belong to their owners and appear only to identify the product being described. Vendor statements are labelled as the vendor's. Rankings are the output of the published method at the edition date. They change when a sourced fact changes, and the read date on a page shows when it was last checked. The site may change or move without notice. Contact the editor at editor@aiendpointsecurityranked.com with questions. ## Related pages - [About the index](/about/) - [Editorial policy](/editorial-policy/) - [Corrections and tips](/corrections/) - [Contact the editor](/contact/) - [Privacy](/privacy/) ## Sources - Method: https://aiendpointsecurityranked.com/method/ (read 12 Nov 2026) Last reviewed 12 Nov 2026. Canonical: https://aiendpointsecurityranked.com/terms/