This “AI” “revolution” is going great. People are handing remote-code-execution-tools, that pull in insecure instructions by default, their full corporate credentials. By the millions. I find it hard to call that a backdoor - it’s a front door.
https://brightbean.xyz/blog/mcp-backdoor-zero-trust-architecture-security/

MCP Is the Backdoor Your Zero-Trust Architecture Missed
The Model Context Protocol connects AI agents to enterprise tools — but it ships without authentication, authorization, or audit trails. With 7,000+ exposed servers and a growing list of CVEs, MCP has become the blind spot in your zero-trust perimeter. Here's what happened, what's at stake, and how to lock it down.





