SAFE-MCP - A Security Framework for AI and Model Context Protocol

Explore a comprehensive security framework designed to protect AI systems using the Model Context Protocol (MCP) in this 22-minute conference talk. Learn about SAFE-MCP, a community-driven security framework that addresses emerging threats in AI tool integration by mapping 65 attack techniques across 14 tactics based on real-world security research from 2024-2025. Discover how attackers exploit MCP through various methods including Tool Poisoning, where hidden instructions in tool descriptions compromise AI behavior, Command Injection affecting 43% of MCP servers according to Equixly research, and novel attacks like MCP Rug Pull. Examine practical case studies from incidents at major organizations and research findings from Invariant Labs, Equixly, and Backslash Security to understand the current threat landscape. Gain actionable intelligence for securing AI deployments through practical detection and mitigation strategies, learn to identify vulnerabilities in AI systems, implement effective security controls, and understand how to contribute to this evolving community-driven security framework for protecting AI+MCP implementations.