Claude Mythos AI Finds 10,000+ Software Vulnerabilities

Anthropic's Claude Mythos AI has identified over 10,000 severe software vulnerabilities. This breakthrough came through "Project Glasswing," a new cybersecurity initiative. The project gave early access to about 50 security partners, including banks and software vendors. The AI system uncovered 6,202 high or critical vulnerabilities in over 1,000 open-source software projects used globally. After human validation, 1,726 legitimate vulnerabilities were confirmed, with 1,094 being high or critical severity. This shows an unprecedented scale for AI-assisted research in a short time. The Claude Mythos Preview can independently review large software repositories, identify weaknesses, and prioritize severity. It can even reason about how attackers could exploit these flaws. Industry researchers found the system dramatically more capable than earlier AI models. This signals that advanced AI is rapidly becoming a key player in cyber defense. This development could significantly change how we protect digital infrastructure.