Anthropic has unveiled Claude Mythos Preview, its most advanced AI model capable of autonomously discovering zero-day vulnerabilities and generating exploit code. The model is currently accessible only to 12 major companies, including Amazon, Apple, Microsoft, and Google, as part of a controlled release due to its powerful capabilities. Anthropic has also initiated Project Glasswing to encourage the use of Mythos for cybersecurity defense. Claude Mythos has demonstrated significant improvements over previous models, achieving a perfect score on the Cybench cybersecurity benchmark and discovering long-standing vulnerabilities in systems like OpenBSD and FFmpeg. Despite concerns about its potential misuse, Anthropic emphasizes the model's role in enhancing cybersecurity by identifying vulnerabilities and guiding remediation efforts.