Question

A global enterprise is deploying autonomous AI agents built on Azure OpenAI and Copilot Studio to automate finance workflows. Each agent must call Microsoft Graph and internal APIs on its own behalf (no signed-in user), and the security architect must ensure the agents follow least-privilege, are individually attributable in audit logs, and can be revoked instantly if compromised. Human owners must periodically confirm each agent still needs its access. Which identity and governance design best meets these requirements?

Accepted Answer

Assign each agent a dedicated Entra ID application with its own service principal, grant only the specific Graph application permissions needed, use managed identities or workload identity federation for credentials, and place the agents' access under Entra ID access reviews and PIM for Groups.

Answer

Have all agents share a single high-privilege service principal with Directory.ReadWrite.All so operations succeed reliably, and rely on Conditional Access to block risky sign-ins.

Answer

Run all agents under the identity of a dedicated human service account with a strong password stored in Azure Key Vault, and require MFA at each API call.

Answer

Grant the agents delegated permissions tied to the finance manager's user account so that all agent actions inherit that user's existing entitlements and audit trail.