Anthropic's Most Dangerous Model Just Got Accessed by People Who Weren't Supposed to Have It

Anthropic built a model so dangerous they refused to release it publicly. Then a Discord group got in anyway. The Model They Wouldn't Ship Claude Mythos Preview is Anthropic's most capable model to date for coding and agentic tasks. Anthropic But it was never meant to reach the public. During testing, Mythos improved to the point where it mostly saturated existing cybersecurity benchmarks, prompting Anthropic to shift focus to novel real-world security tasks — specifically zero-day vulnerabiliti