Zero Trust for Autonomous Agents: Isolating AI Workloads on Kubernetes
summary
Conference claim that half a million AI agents will become digital coworkers leads to a concrete Zero Trust blueprint: map cell = Kubernetes namespace, enforce north/south/east/west gates with Envoy and network policies, and require agents as first‑class identities using OIDC/OAuth2 and token exchange (including CIBA). Provider credentials stay in the gate, not the agent, calls are metered and PII‑filtered, every hop is audited into a single trace ID, and two reference OSS projects are shown: Open Coryo and Thunder ID.
AI-written summary · about
chapters (7)
- 0:00Autonomous agents as coworkers
- 3:51Zero trust for AI agents
- 8:45Cell-based architecture in Kubernetes
- 12:54Consent and human-in-the-loop flows
- 18:12Isolation and identity layers
- 21:00Q&A on security and architecture
- 39:25Dynamic agent permissions and policies
related talks
30mZero-Trust Traffic Governance for Kubernetes AI Agent Sandboxes
40mIn AI We Trust? Securing the Future, One Agent at a Time
25mWhat We Learned Securing AI Agents at Scale on Multi-Tenant Kubernetes Clusters
23mZero-Trust Architecture for Agentic AI: Securing Multi-User Access and Third-Party Integrations
22mBuilding Secure AI Platform Infrastructure: A Zero Trust Framework
38mToward Zero Trust with AI
45mZero Trusting as a True Cloud Native Dev
35mDemystifying Zero-Trust for Cloud Native Technologies