Anthropic leaks 'Claude Mythos' — new model with dramatically better performance

A configuration error in Anthropic's CMS has revealed the existence of Claude Mythos, the company's most powerful AI model to date. The leak exposed nearly 3,000 unpublished internal assets, including a draft launch blog post.

Mythos introduces a new fourth model tier called Capybara, positioned above the current Opus tier. According to leaked documents, the model scores "dramatically higher" than Claude 4.6 Opus on programming, academic reasoning, and cybersecurity tasks.

Most striking are the model's cybersecurity capabilities. Anthropic internally warns that Mythos is "currently far ahead of any other AI model in cyber capabilities" and "presages an upcoming wave of models that can exploit vulnerabilities in ways that far outpace the efforts of defenders."

As a result, Anthropic plans a cautious rollout, providing early access primarily to organizations focused on cyber defense. The news caused a significant drop in cybersecurity stocks on March 27, 2026, with CrowdStrike and Palo Alto Networks among the hardest hit.

Anthropic has confirmed that Mythos training is complete and the model is being piloted with select customers. The company emphasizes it is expensive to operate and not yet ready for general release.

Sources: SiliconAngle, Fortune, Mashable, The Decoder