Anthropic has officially confirmed the existence of Claude Mythos, a model it describes as the most capable it has ever built — and one it has no plans to release publicly.
Project Glasswing
Access to Claude Mythos is being managed through Project Glasswing, a restricted program for approximately 50 partner organizations. Partners are primarily focused on two domains: advanced cybersecurity research and complex multi-step reasoning tasks that exceed the capabilities of publicly available models.
Why not release it?
Anthropic CEO Dario Amodei stated that Mythos's capabilities in certain domains — particularly autonomous action and persuasion — cross thresholds that require controlled deployment. The company says it is using learnings from Glasswing to develop safety protocols before any broader release.
What we know about its capabilities
Anthropic has not published benchmark results for Mythos. Partner organizations have described it as capable of sustained autonomous operation across complex multi-day tasks — a qualitative leap beyond what Claude Opus 4.7 can achieve. One cybersecurity partner described it as able to conduct end-to-end penetration testing with minimal human direction.