200,000 MCP servers have a critical flaw Anthropic won't fix

追加: 2026-05-16

[00:00:00]200,000 AI servers just got exposed by a command execution flaw, >> [music] >> and Anthropic is calling it a feature, not a bug. This matters because every major AI company adopted this protocol.

[00:00:13]OpenAI, Google DeepMind, thousands of developers, they're all running the same vulnerable code. Here's what happened.

[00:00:21]Four security researchers found that MCP's default transport lets attackers execute arbitrary commands on any connected system. [music] The architecture itself is the problem, not some implementation mistake.

[00:00:36]Anthropic's response? [music] They say this is by design, that it's the developers' responsibility to secure their own setup. But millions of installations don't know this risk exists yet. The real play here is speed.

[00:00:52]Patches are coming. Companies are scrambling. Whoever moves first controls the narrative around AI security for the next decade. Follow. This story is moving fast.