Anthropic's next-generation flagship model has a name — Claude Mythos — and it leaked by accident. A misconfigured content management system exposed roughly 3,000 unpublished internal documents, including a draft blog post announcing the model.

What Got Leaked

The draft describes Claude Mythos (codenamed Capybara internally) as a model in an entirely new tier above Opus — the company's current most capable model. Anthropic reportedly characterizes it as "by far the most powerful system we have ever developed," with major leaps in coding, reasoning, and especially cybersecurity capabilities.

The company says it is "currently far ahead of any other AI model in cyber capabilities," a framing the draft treats as both a selling point and a concern.

Cautious Rollout Planned

Anthropic says it plans to roll out Mythos slowly, with initial access limited to select cybersecurity firms — the intent being to help defenders harden their systems before the model becomes more broadly available. The draft explicitly acknowledges that the model "poses unprecedented cybersecurity risks" and that the team has unresolved concerns about unintended consequences at release.

The leak comes just days after a federal judge ruled in Anthropic's favor in its lawsuit against the Pentagon, which had designated the company a national security supply-chain risk.

Industry Context

The accidental leak follows weeks of speculation about Claude 5. Both OpenAI and Anthropic have recently acknowledged that upcoming models represent capability jumps significant enough to warrant unusual caution. OpenAI's rumored "Spud" model has drawn similar attention, with the company renaming one of its internal groups "AGI Deployment."

Anthropic has not yet formally confirmed the details of the leaked documents.