Review AI agent identities and their access
Permafrost shows collected Entra agent identities, their observed access and available risk scores. Use this evidence to review agent access and inform the control areas described below.
This page is about how Permafrost helps you govern your AI agents. For Permafrost’s own AI-management-system posture (ISO 42001) and its certification status, see certification status. We map only what the product genuinely supports; contributing to a framework’s control area is not the same as certifying against it.
Framework crosswalks
EU AI Act
The EU AI Act expects providers and deployers of higher-risk AI systems to maintain oversight, logging, and access control over the system and the identities that operate it.
- Human oversight & access control (Art. 14, Art. 26)Agent identity inventory lists tracked Entra agent identities from collected data and shows their blueprints and lifecycle sponsors where available.
- Risk management & technical robustness (Art. 9, Art. 15)Permafrost flags high-risk inheritable permissions declared on collected agent blueprints.
- Record-keeping & traceability (Art. 12)Per-agent blast radius shows confirmed and potential access paths in the collected access graph.
NIST AI RMF
The NIST AI Risk Management Framework organizes AI risk work into Govern, Map, Measure, and Manage. The identity layer for AI agents maps cleanly onto each.
- Map — context & inventoryThe inventory helps you review the Entra agent identities and lifecycle sponsors Permafrost has collected.
- Measure — risk analysisThe principal risk score combines available posture signals and findings to help prioritize identities for review.
- Manage — risk treatmentSign-in findings identify agents for review when collected evidence shows no recent sign-in. An analyst decides whether to disable an identity.
CSA AICM
The Cloud Security Alliance AI Controls Matrix extends cloud-control thinking to AI systems, with identity and access management among its control domains.
- Identity & access management for AIReview collected Entra agent identities alongside their available risk scores.
- Least privilege & entitlement hygieneBlueprint inheritable-permission risk and per-agent blast radius surface the entitlement excess and propagation paths the matrix expects you to control.
- Lifecycle managementUse agent sign-in findings to review whether standing access is still needed.
OWASP AI Exchange
The OWASP AI Exchange catalogs AI security threats and controls, including excessive agency and over-permissioned autonomous agents — the failure modes Permafrost is built to surface.
- Excessive agency / over-permissioned agentsReview access paths in the collected graph and high-risk permissions declared on agent blueprints.
- Least-privilege enforcementUse the principal risk score to prioritize identities, then review the underlying findings and evidence.
- Monitoring & decommissioningReview agents with no recent sign-in in the collected evidence and verify workload dependencies before removing access.
Outcome first: fewer over-permissioned agents, less standing access, and audit evidence you can hand to a reviewer.
