Agentic AI systems are increasingly expected to operate across organizational boundaries, where distinct trust domains govern identity and access control. This session explores how to design secure and scalable authorization flows for agentic AI collaboration using open standards such as OAuth 2.1, Model Context Protocol (MCP), and Agent-to-Agent (A2A).
Starting with a single-domain setup, Yoshiyuki Tabata walks through the implementation of authorization code flow, token introspection, and resource protection. He then introduces the concept of multi-domain environments, where tokens issued in one domain cannot be reused in another. To address this, he explains how OAuth 2.0 Token Exchange (RFC 8693) enables agents to securely obtain new tokens for downstream services in other domains.
The session concludes with a demonstration using Keycloak, showcasing federated authorization in action. Attendees will gain practical insights into building secure, interoperable agent infrastructures across trust boundaries.