⚡Lightning Talk: Securing AI Agents: Applying Service Mesh Patterns To MCP
Lightningat Agentics Day: MCP + Agents Europe 2026 · on Monday, 2026-03-23· at 17:30· Hall 7 | Room A·
summary
Claims that MCP deployments lack mutual TLS and per-tool authorization and demonstrates applying service mesh primitives to close those gaps using Istio (Envoy sidecars, Istiod-issued identities) and Linkerd (always-on mTLS, 24‑hour cert rotation). Maps concrete attack vectors—tool poisoning, rug pull, cross-server shadowing, data exfiltration—to defenses: mTLS stops impersonation, SPIFFE-style workload identities prove provenance, and authorization policies block unauthorized tool calls. A live demo shows an authorization policy denying a malicious server while allowing a trusted server to access repository, run commands, and query the database.
AI-written summary · about
related talks
06mLightning Talk: Securing Your Systems with Service Mesh
44mMCP Security: Keep Your AI Agents from Spilling the Tea
27mSecuring the AI Revolution - [...]
29mMCP Security: Keep Your AI Agents from Spilling the Tea
11mLightning Talk: Your Most Privileged User Isn't Human: The Agent Security Blinds...
15mWhy Security Matters: The Risks of Agentic AI and How to Mitigate Them
27mService Mess to Service Mesh
32mIstio: Sailing to a Secure Services Mesh