Is your AI agent infrastructure secure?
As AI agents begin to exchange model context and coordinate across systems, secure interaction is no longer optional—it’s essential. To bring structure to these interactions, protocols like Model Context Protocol (MCP) and Agent-to-Agent (A2A) have emerged, offering standardized ways for agents to communicate.
Adopting these protocols introduces new responsibilities. Developers must implement authentication and authorization (AuthN/AuthZ) mechanisms that comply with MCP and A2A while remaining practical for real-world deployment.
In this session, Yoshiyuki Tabata shares best practices for designing AuthN/AuthZ and shows how to apply key principles from the CNCF IAM whitepaper to AI agent infrastructure—such as OAuth-based API access, P*P architecture for authorization, and workload authentication. The session includes a demo of secure AuthZ for an MCP server using Keycloak, illustrating how these practices apply in real-world agent interactions.