Claude Mythos AI Finds 10,000+ Zero-Days in Project Glasswing

Anthropic's Project Glasswing, a new cybersecurity initiative, has uncovered over 10,000 high and critical severity zero-day vulnerabilities in its first month. This project uses the unreleased Claude Mythos Preview AI model to secure critical infrastructure. The model worked with over 50 technology organizations, including Microsoft, Apple, and Google, to scan critical software systems. Cloudflare reported finding 2,000 bugs, and the model's false-positive rate is lower than human testers. The UK's AI Security Institute confirmed the model's ability to solve multistep cyberattack simulations. Mozilla used it to patch 271 vulnerabilities in Firefox 150. Anthropic is keeping Mythos from public release due to its autonomous exploit capabilities. It also scanned over 1,000 open-source projects, finding a critical flaw in the wolfSSL cryptography library. The model even engineered an exploit for this vulnerability. The sheer volume of discoveries shows a major weakness in the software industry. Human capacity to patch vulnerabilities cannot keep up with AI-driven discovery. This impacts everyone who uses software, highlighting a growing challenge in cybersecurity.