AI has another security problem We all know about how MCP has fundamental flaws in how it handles authorization, how OpenClaw and its relatives open up local machines to exploits and the whole world of LLMs is a security nightmare𐠒. And we also are now hearing that Mythos, Anthropic’s latest model, can exploit all of the security flaws in all of the software ever and it’s so impressive that it was released ahead of time to everyone important so they could run it themselves and protect themsevelves from the oncoming onslaught of hackers equipped with it.