Anthropic's Claude Mythos Leaked: A New Tier Above Opus

Anthropic accidentally left draft blog posts in a publicly accessible CMS cache — and cybersecurity researchers found them. The posts reveal Claude Mythos, described internally as "by far the most powerful AI model we've ever developed." A second name, Capybara, appears to refer to a follow-up version.

Mythos sits in a brand-new tier above Opus — previously the top of Anthropic's lineup. The draft claims it scores dramatically better on coding, reasoning, and cybersecurity benchmarks. Anthropic confirmed the leak to Fortune, calling it "a step change," and said it's currently in early-access testing with a small group of customers.

The draft doesn't just hype the model — it flags serious risk. Anthropic warns Mythos "presages an upcoming wave of models that can exploit vulnerabilities in ways that far outpace the efforts of defenders." That's an unusually candid admission about a product not yet shipped.

Why it matters: A frontier AI lab is publicly acknowledging its own upcoming model could outrun cybersecurity defenses — a rare and sobering signal about where capable AI is heading.