Anthropic's Claude Mythos Preview, a new general-purpose language model, has demonstrated exceptional capabilities in cybersecurity, specifically in identifying and exploiting zero-day vulnerabilities across major operating systems and web browsers. Through their Project Glasswing initiative, they have used Mythos Preview to autonomously identify and develop exploits for long-standing security flaws, such as a 27-year-old OpenBSD bug and a 17-year-old FreeBSD remote code execution vulnerability, showcasing a significant advancement in AI-driven security tools that may transform how the industry defends against cyberattacks.
