AI vulnerability discovery

Anthropic’s Claude Mythos AI model preview outperforms every prior Claude model on cybersecurity benchmarks and has already found thousands of high-severity vulnerabilities across every major OS and browser. It is restricted to a closed coalition of 11 partners including AWS, Microsoft, Google, and Apple under Project Glasswing. Here is what it can do, why access is locked down, and what every CISO needs to do before the end of 2026.