FusionAuth Adds OAuth Controls for Customer-Facing AI Agents

FusionAuth has released version 1.70 of its customer identity platform with controls intended to help application providers authorize AI agents without granting them unrestricted access to user accounts.

The update adds support for Client ID Metadata Documents, a mechanism that allows compatible AI applications to register under policies set by the service provider. It also implements OAuth 2.0 token exchange under RFC 8693, allowing a service to request a constrained token for access to another system on a customer's behalf.

Administrators can restrict the permissions granted during an exchange, while FusionAuth records the request and the policy decision. The company has also added an optional intent statement that can be stored with the exchange record, giving developers and auditors additional context about what an agent said it planned to do.

“Customers will hand agents bigger jobs once they trust the company on the other end,” said Brian Bell, CEO of FusionAuth. “That trust starts with a clear record of who an agent acted for, what it was allowed to do, and why, all of it kept on infrastructure the company controls. AI can assist, but identity decides.”

The controls are relevant to companies exposing tools through Model Context Protocol servers or other agent interfaces. Authentication can establish which user approved an agent, but authorization determines the specific resources and actions available after that approval. Token exchange can reduce risk when it produces narrowly scoped, short-lived credentials instead of passing a user's original token through multiple services.

Client ID Metadata Document support and OAuth token exchange are available in FusionAuth 1.70 on the Enterprise plan. Organizations should still test how clients are validated, how intent records are protected from manipulation, and whether downstream services enforce the scopes attached to exchanged tokens.