Google made three-legged OAuth authentication through Agent Identity auth manager generally available for Gemini7.5 out of 10 Enterprise A2A agents on November 13, 2026. The integration lets supported agents access downstream tools on behalf of the signed-in user, with Gemini Enterprise managing the consent prompt and resuming the conversation afterward. The dated release notes identify agents built with Google’s Agent Development Kit and deployed on Agent Runtime as the supported configuration.
Consent becomes part of the agent conversation
An agent may need a separate authorization before it can use an MCP server or third-party API. Under the documented flow, the agent signals that credentials are required, and Gemini Enterprise opens the provider’s authorization page for the user.
After consent, Gemini Enterprise validates the session, completes credential setup with the auth manager, and returns control to the agent. Developers therefore do not need to build a separate client interface for that consent-and-resume sequence. They still need to configure the OAuth application, auth provider, and agent correctly. Google’s setup guide describes those requirements.
A practical alternative to shared access
Consider an internal assistant asked to inspect a user’s project tickets. User delegation allows the tool request to use that person’s authorized access instead of relying solely on a broadly privileged credential shared by the application.
Google’s architecture separates the human user, the agent’s own identity, and the delegated authority used for a particular tool call. Outbound requests must satisfy both the agent’s access policies and the user’s permissions and consented OAuth scopes. Audit records capture both identities. That makes the distinction useful when investigating who initiated an action and which workload performed it. The identity architecture documentation explains these boundaries.
General availability has a specific scope
The November 13 announcement applies to the authentication integration. Google’s broader A2A registration guide still carries a Preview label, so the update should not be read as making every surrounding A2A capability generally available.
There are implementation constraints too. Gemini Enterprise uses A2A’s version 0.3 streaming mechanism; newer agents need compatible interfaces. Developers must also use the documented credential-interrupt handling so an authorization request reaches the user correctly.
For teams building enterprise agents, the change reduces custom consent-interface work. Permissions, protocol compatibility, and deployment configuration remain part of the integration.