Architecting Secure Agentic Workflows on Kubernetes: A Financial Sector…
summary
Reference architecture uses MCP Gateway, SPIFFE/SPIRE and a transparent proxy (ROSOctl) with Keycloak-backed token exchange to give agents short‑lived, scoped JWTs; MLflow records traces and an LLM-based judge flags prompt‑injection. A financial agent fetches market data and executes simulated trades, a prompt‑injection causes unauthorized outbound tool calls, MLflow alerts after the fact, and the proxy’s real‑time guardrail successfully blocks malicious tool responses; testing runs 10–20k harnessed paths to measure fat‑tail failures.
AI-written summary · about
chapters (6)
- 0:00Introduction and agenda
- 1:17Agentic security challenges
- 5:15System architecture
- 9:39Observability and evaluation
- 16:05Demonstration
- 27:30Conclusion and defense
related talks
40mArchitecting Secure Agentic Workflows on Kubernetes: A Financial Sector...
27mKeeping Your Agents in Check: Layered Security for Agentic Platforms in Production
25mWho Let the Agents Out? Securing AI Workflows the Right Way
36mSecuring Agentic Platforms with Zero-Trust Workload Identities
14mArchitecting Secure and Scalable Integration Platforms for FinTech
39mSecure Authorization for Agentic AI in Multi-Domain Environments
11mBuilding Secure Backend Services with AI Agents
27mMCPIdentity: Keyless MCP Agent Authentication Patterns on Kubernetes With Keycloak…