Anthropic's Secret AI Model

Anthropic accidentally left their unreleased model announcement in a public data store. The internet found it before the press release did.

Claude Mythos — internally codenamed Capybara — is a new tier above Opus 4.6, currently Anthropic's most powerful model. The leaked draft claims step-change improvements on coding, academic reasoning, and cybersecurity. That last one is exactly why Anthropic is being careful. A model this capable in security doesn't just help defenders — it could help attackers move faster than anyone can respond. Their words, not ours.

Public release unconfirmed.
They built it, got nervous, and then accidentally told the entire internet it exists.