Securing Non-Human Identities: A Defense-in-Depth Blueprint for AI Agents
summary
At LINE/Yahoo! JAPAN scale the system serves tens of millions of users across more than 10,000 Kubernetes nodes and fixes the resurrected confused deputy by propagating human intent cryptographically through ID-JAG and RFC 8693 token exchange. Keycloak (26.2/26.7) and Athens issue audience‑pinned ID‑JAGs with minute TTLs, exchanged via JWT bearer grants so privileges narrow at each hop. A live demo uses three curl requests to show ID token → ID‑JAG → short‑lived access token and the calls succeed.
AI-written summary · about
chapters (8)
- 0:00Securing AI agent identities
- 1:02Confused deputy attack with AI
- 3:53Identity chaining defeats attack
- 8:35RFC 8693 and ID Jack
- 14:21Athens workload identity
- 18:01Live demo identity chaining
- 23:033 steps to implement
- 24:07Closing resources
related talks
17mDon't give your agent the master key
27mGive Agents the Right Keys - Identity Assertion for Enterprise APIs
25mIdentities and Authentication for Your Agents With Keycloak
40mIn AI We Trust? Securing the Future, One Agent at a Time
35mIdentity and Access Management for the Decade Ahead
01h 22mIdentity for AI Agents
09mNavigating the Identity Abyss in the AI‑Native Era
20mCIAM for AI: Authn/Authz for Agents