AI Agent Identity and On-Behalf-Of Authorization
AI agent on-behalf-of authorization: delegated identity, OAuth token exchange, actor claims, credential downscoping, runtime policy, and audit trails.
Read guideAI Security
Agentic identity guides
Authentication, delegation, workload identity, and action-level authorization for AI agents.
What these guides cover
Agentic identity separates the people, software actors, workloads, OAuth clients, and target services involved when an AI agent acts on behalf of someone else. A user token alone hides which agent performed an action, while a workload identity alone does not carry user consent or permission to a specific resource. Secure delegation preserves both subject and actor, downscopes credentials for each audience, limits delegation-chain growth, and rechecks authorization immediately before a consequential action. Identity establishes who may participate; runtime policy still decides whether this operation, with these arguments and consequences, should proceed now. Audit records must preserve that distinction across every downstream service.
Risks
- Agents sharing broad user bearer tokens
- Actor identity lost across service boundaries
- Scope or authority increasing through delegation chains
- Valid delegation mistaken for current user intent
Control priorities
- Separate subject, actor, workload, client, and tenant
- Audience-restricted, short-lived, downscoped credentials
- Object and action authorization at the tool boundary
- Revocation design and complete delegation audit trails
Recommended reading order
Use the on-behalf-of guide as the implementation path from principal inventory through OAuth token exchange, actor claims, multi-agent delegation, runtime authorization, token lifecycle, audit evidence, and tool selection.