MCP for agent-to-agent comms may be the riskiest protocol you've never heard of

A new communication protocol for AI systems has security vulnerabilities that could let malicious instructions spread between AI agents. This matters to people using AI tools and companies deploying them.