Blog Post

Microsoft Security Community Blog
7 MIN READ

Authorization and Governance for AI Agents: Runtime Authorization Beyond Identity at Scale

ashwinijwaghmare7's avatar
ashwinijwaghmare7
Icon for Microsoft rankMicrosoft
Apr 07, 2026

This architecture is intended for organizations building AI agents that invoke tools or protected APIs—evolving from securing agent identity to governing autonomous execution at enterprise scale using a Microsoft Entra‑protected Authorization Fabric.

Designing Authorization‑Aware AI Agents at Scale Enforcing Runtime RBAC + ABAC with Approval Injection (JIT) Microsoft Entra Agent Identity enables organizations to govern and manage AI agent ide...
Updated Apr 07, 2026
Version 1.0
audit
azure
azure security center
cloud security
compliance
compliance management
compliance manager
data loss prevention
email security
enterprise security
BlakeGoforth's avatar
BlakeGoforth
Copper Contributor
Aug 08, 2026

This is very close to the runtime governance problem I have been exploring. I’m building a pilot-grade open reference project called SMERC: recoverability-aware runtime permission infrastructure for AI-agent actions. The narrow question SMERC adds beside identity, RBAC/ABAC, approval injection, and policy decisioning is:

If this agent action is technically authorized, is it recoverable enough to execute now?

 

SMERC evaluates signals such as reversibility, containment strength, rollback latency, evidence validity, anomaly pressure, impact scope, and executor fitness. It then returns a replayable posture: ALLOW, THROTTLE, FREEZE, DENY, or ESCALATE. It does not replace Microsoft Entra, OPA, IAM, AI gateways, SIEM, EDR, approval workflows, or policy engines. The goal is to add a recoverability checkpoint before high-impact agent actions execute.

 

Public review site: https://admirable-sorbet-9986d5.netlify.app/

GitHub: https://github.com/KingsMtn/SMERC-Runtime-Permission-Layer

 

I would value technical feedback from this community on whether recoverability scoring is a useful additional runtime governance signal, especially for GitHub Actions, deployment workflows, cloud operations, and security-response automation. Current boundary: SMERC is pilot-grade software with public code, tests, and replayable evidence. It is not production-certified or customer-proven yet.

ashwinijwaghmare7's avatar
ashwinijwaghmare7
Icon for Microsoft rankMicrosoft
Aug 26, 2026

BlakeGoforth​ 

Thank you for sharing this. I like the distinction between authorization ("Should this action be allowed?") and recoverability ("If something goes wrong, can we safely recover?").

Recoverability signals such as reversibility, rollback latency, containment, and impact scope could be valuable additional inputs for high-impact agent actions. The interesting architectural question is whether these become part of the authorization decision itself or operate as a complementary governance checkpoint.

Appreciate you sharing the project. I'm sure the community will find it an interesting contribution to the broader discussion around safe and scalable AI-agent operations.