Assessing Claude Mythos Preview's cybersecurity capabilities (red.anthropic.com) AI

Anthropic says its Claude Mythos Preview model shows “next-generation” strength in cybersecurity research, including finding and exploiting zero-day vulnerabilities across major operating systems and browsers. In testing under Project Glasswing, the company reports Mythos Preview can construct complex exploits (including sandbox-escaping and privilege-escalation chains) and turn known or newly discovered vulnerabilities into working attacks. The post details their evaluation approach and notes that most reported findings remain unpatched, so they provide limited disclosure while urging coordinated defensive action from the industry.