Anthropic released Claude Mythos Preview, a frontier model that autonomously finds and exploits zero-day vulnerabilities, and launched Project Glasswing to deploy it defensively. The model discovered thousands of high-severity flaws in major operating systems and browsers—including bugs up to 27 years old—writing exploits in hours that human pentesters estimated would take weeks. Anthropic is restricting public access; 12 founding partners (Amazon, Apple, Google, Microsoft, CrowdStrike, and others) receive controlled access under the initiative, backed by $100 million in model credits and $4 million in open-source security donations.