Anthropic is about to release Claude Mythos preview, focused on computer security

Anthropic seems ready to release a preview of Claude Mythos. Sources say the model option 'claude-mythos-1-preview' briefly appeared in public versions of Claude Code and Claude Security, then was removed, suggesting Anthropic is likely in final testing.

Claude Mythos is designed as a cutting-edge model for computer security tasks, with notable improvements in code reasoning and autonomous execution over the current flagship Opus 4.7. However, Anthropic has warned that Mythos can automatically develop professional-level cyberattack methods, posing serious risks to global digital infrastructure if released.

Meanwhile, Anthropic is pushing forward with a new project called 'Glasswing,' partnering with other companies to protect critical software systems worldwide from AI-powered exploits. The initiative uses the unreleased Claude Mythos Preview and has already helped up to 50 partner organizations.