Securing Agentic Platforms with Zero-Trust Workload Identities
summary
Shows a working pattern: workload attestation with SPIFFE/SPIRE agents on every node, Keycloak as the OIDC/OAuth2 broker (new Keycloak 26.2 SPIFFE identity provider), and Vault/OpenBō for secrets. Implements per-hop short-lived OIDC tokens and token-exchange so user tokens never propagate, enforces attribute-based authorization and trust-domain boundaries, and demonstrates mitigation of MCP risks (tool poisoning, RAG exfiltration). Components are integrated and automated (Spire controller for k8s) and the design reduces token theft and cross-domain impersonation.
AI-written summary · about
chapters (8)
- 0:00Agentic AI workload identities
- 3:54Securing unpredictable AI agents
- 7:11Agent identity anti-patterns
- 12:35Zero-trust for agent workloads
- 18:14Workload identity security tools
- 23:53MCP security pitfalls
- 27:44Future of agent identity
- 30:13Q&A
related talks
32mA Platform-Agnostic Approach to Securing Workloads
30mAdvancing Machine to Machine Workflows with the SPIFFE Identity Provider
37mZero Trust Workload Identity in Kubernetes
25mBridging the Gap Between Human and Workload Identity: SPIFFE Support in Keycloak
41mZero-Trust Networks: Identity, AuthN, AuthZ, Audit
34mThrow Away Your Passwords: Trusting Workload Identity
34mThrow Away Your Passwords: Trusting Workload Identity
39mOn Establish a Production Zero Trust Architecture