Secure Authorization for Agentic AI in Multi-Domain Environments
Talkat Open Source Summit Japan + AI_dev: Open Source GenAI & ML Summit + Automotive Linux Summit 2025 · 2025-12-08..12-10·
summary
Demonstrates a concrete secure authorization design that spans two Kubernetes clusters with two Keycloak instances and sample MCP servers, showing dynamic client registration, token introspection, and an actual token exchange where a green token from domain A is swapped for a pink token in domain B using OAuth 2.0 Token Exchange (RFC 8693) plus a JOT-based identity grant so identity context propagates. Uses MCP and A2A protocols, implements sidecar introspection and broker components, and the demo works end-to-end as captured in the network traces.
AI-written summary · about
chapters (6)
- 0:00Secure AI agent auth intro
- 1:45MCP and A2A auth
- 6:04Single domain auth flows
- 11:21Multi-domain auth challenges
- 16:20Federated auth demo with Keycloak
- 35:27Summary
related talks
32mAI Agents with Keycloak in MCP
27mSimple Control: Managing Multi-Domain MCP Access Via Keycloak IdP With ID-JAG
23mIdentity Best Practices to Secure & Scale Agentic AI / MCP Deployments
25mModel Context Protocol(MCP) Authorization for Enterprise Use Case With Keycloak
36mSecuring Agentic Platforms with Zero-Trust Workload Identities
27mKeeping Your Agents in Check: Layered Security for Agentic Platforms in Production
44mMCP Security: Keep Your AI Agents from Spilling the Tea
29mMCP Security: Keep Your AI Agents from Spilling the Tea