Anthropic Mythos 1 for Claude Code and Security
Key Questions
What is Mythos 1?
Mythos 1 is Anthropic's new model targeted at Claude Code and security applications. It uses the MSM technique to improve alignment and safety.
What results has Mythos 1 shown on bugs?
Glasswing testing with Mythos 1 identified over 10,000 critical bugs. These findings support its move toward public release.
How does the MSM technique affect performance?
MSM trains models on documents about the spec rather than the spec alone. This reduced misalignment rates from 68% to 5%, enhancing safety implications.
Mythos 1 targets Claude Code/security with MSM technique slashing misalignment 68% to 5%. Glasswing results show 10k+ critical bugs found; moving closer to public release with safety implications.
Sources (2)
Updated May 24, 2026