MCP for agent-to-agent comms may be the riskiest protocol you've never heard of
2026-10-05 · Ars Technica
The Model Context Protocol (MCP), designed to let AI agents talk to each other, turns out to have trust gaps that let malicious prompts hop from one agent to another like a digital plague. Because what could possibly go wrong with giving AI agents a group chat where nobody verifies who's saying what? At least when humans spread misinformation, they have to type it themselves.