How to Inventory and Govern AI Agent Identities in Entra ID

Retrospective: this article looks back at events from May 2025, written in 2026 with the benefit of hindsight.

AI agents are a fast-growing category of identities with access to company data and systems. Here is how to inventory and govern them in Microsoft Entra ID.

Step 1: Find your agents

Look in several places:

Entra Agent ID inventory (agents registered with agent identities).
Microsoft 365 admin center → Agents (Copilot agents and integrated apps).
Copilot Studio environments (Power Platform admin center).
Azure AI Foundry projects.
App registrations and service principals used by custom agents (often named "bot," "agent," "assistant," "GPT").
Defender for Cloud AI security posture management for AI workloads in Azure and other clouds.

Step 2: Assign owners

Every agent needs a business owner and a technical owner. Unowned agents should be disabled after notice.

Step 3: Review permissions

For each agent, list:

Data it can read (mailboxes, sites, databases).
Actions it can take (send email, update records, call APIs, make payments).
Whether it runs with its own identity or a user's delegated permissions.

Reduce to the minimum required. Prefer agent-specific identities over shared or user credentials.

Step 4: Apply policies

Conditional Access and governance policies for agent identities where supported.
Access reviews for agents with sensitive permissions.
Purview DLP and sensitivity labels to limit what agents can process.

Step 5: Require human approval for high-impact actions

Payments, external email, deletion and permission changes should require human confirmation.

Step 6: Monitor

Log agent sign-ins and actions; alert on new permissions and unusual activity.

Verify

Quarterly: agent count, owner coverage, agents with high-impact permissions.