QuestionQ56

VerifiedSecuring and governing agentic workflows

You are configuring a custom AI agent deployed in Gemini Enterprise and the Conversational Analytics API. The agent calls downstream enterprise tool APIs running in a separate Google Cloud project to retrieve restricted business analytics on behalf of the logged-in user. Currently, custom tool logic handles unmanaged raw OAuth 2.0 access tokens from execution context state, causing authorization failures during cross-project API calls when tokens expire and exposing raw long-lived credentials in custom tool code.

You need to establish a secure, managed identity propagation architecture between Gemini Enterprise and the downstream tool APIs.

What action should you take?

Explanation

A Google OAuth 2.0 access token in the HTTP Authorization: Bearer header authenticates calls to the Conversational Analytics API. The roles/geminidataanalytics.dataAgentUser role permits a principal to chat with a data agent; access to underlying data sources can require additional permissions. (docs.cloud.google.com)

C is the best available choice, but it is incomplete: the header and IAM role alone do not manage token refresh or prevent custom code from handling credentials. A managed identity propagation architecture also needs a managed credential mechanism, such as Agent Identity auth manager with Agent Gateway where supported. (docs.cloud.google.com)

Learn more

Community Discussion

No comments yet. Be the first to start the discussion!