What is Anthropic's Mythos? The leaked AI model that poses 'unprecedented' cybersecurity risks

A leaked draft says Anthropic is building a new model, reportedly called Claude Mythos or Capybara, that may be much more capable than its previous systems.

The model is said to bring major gains in reasoning, coding, and cybersecurity, raising concerns it could enable large-scale cyberattacks.

Anthropic has confirmed the model exists and says it is working to improve efficiency and safety before release.

The leak underscores fears that more powerful AI could automate offensive cyber operations, putting more pressure on companies and governments to strengthen defenses.