E386 - Securing the Agentic Revolution: The New Rules of Model Context Protocol Security

As AI agents move from experimental pilots to production via the Model Context Protocol (MCP), they introduce a fundamental architectural shift where Large Language Models sit at the center of security-critical decisions. This episode unpacks the Coalition for Secure AI’s comprehensive framework, exploring twelve core threat categories that range from novel vectors like tool poisoning and shadow servers to the "confused deputy" problem. Tune in to learn why traditional perimeter defenses are insufficient and how to implement defense-in-depth strategies, including cryptographic identity propagation, hardware-based isolation, and zero-trust validation for AI outputs

https://cisomarketplace.com/blog/ai-agent-security-crisis-mcp-vulnerabilities

...