Program

MCP (Model Context Protocol) is often described as USB-C for AI: a standard way to connect models to tools, data, and systems. As language models become more capable and more connected, MCP makes it easier to build systems that can do much more than just generate text.

That shift is exciting, but it also raises difficult questions about security, safety, and trust. In this talk, I’ll use MCP as a lens to explore what changes when AI systems start acting beyond the chat window, and why that should make us think harder about control, authority and what we’re actually willing to trust these systems with.