Anthropic Just Teased Another Model..

Anthropic has developed Claude Mythos, a highly capable AI model for cybersecurity research under Project Glasswing. Mythos achieves a 72.4% success rate writing exploits for known vulnerabilities, far surpassing Claude Sonnet and Opus. It can autonomously find and chain zero-day vulnerabilities across major OSes, browsers, and even memory-safe codebases written in Rust. Anthropic is not releasing Mythos publicly, instead partnering with major infrastructure companies like Cisco, Nvidia, and Microsoft to use it defensively. The author discusses the asymmetric offense/defense problem, the democratization of security research through AI, and predicts a turbulent transition period before AI ultimately makes software more secure.