How a cavalcade of blunders gave unauthorized users access to Claude Mythos — restricted model accessed by third parties, thanks to knowledge from data breach

Unauthorized individuals have accessed Anthropic's new Mythos cybersecurity-focused AI model, despite the developer locking it down to just a handful of companies. Considering the AI was purposefully designed to find zero-day exploits and offer viable fixes, the breach raises questions about Anthropic's own security, and why Mythos couldn't protect it.