A Data Leak Revealed Anthropic's Most Powerful AI Yet: Claude Mythos

Anthropic is developing a new AI model more capable than anything it has released so far, and the world found out through a data breach.

A cache of nearly 3,000 unpublished internal documents was left in a publicly searchable data store, including what appeared to be a draft blog post announcing the model. Fortune reviewed the materials and reported the findings. After being notified, Anthropic removed public access to the data store.

The leaked draft refers to the model as Claude Mythos, with a new tier called Capybara. Capybara is described as larger and more capable than Anthropic's current Opus models, which have been the company's most powerful to date. "Compared to our previous best model, Claude Opus 4.6, Capybara gets dramatically higher scores on tests of software coding, academic reasoning, and cybersecurity, among others," the draft read.

Anthropic confirmed the model exists. A spokesperson said it "represents a step change" and is "the most capable we've built to date." The company said it is currently being tested with a small group of early access customers.

The document also flagged "unprecedented cybersecurity risks" posed by the new model, a striking admission given how capable it claims to be. Anthropic said it is being deliberately careful about how the model is released, consistent with its safety-first positioning.

The leak was attributed to a "human error" in the configuration of Anthropic's content management system. It came just days before a second breach: Anthropic also accidentally exposed the source code for Claude Code, its popular coding assistant.