Assessing Claude Mythos Preview's cybersecurity capabilities
Why it matters: Mythos Preview found and exploited zero-day vulnerabilities in every major OS and web browser.
- Anthropic announced Claude Mythos Preview, a general-purpose language model with striking capabilities in computer security tasks.
- Mythos Preview can identify and exploit zero-day vulnerabilities in every major operating system and web browser, with some discovered bugs being decades old, including a 27-year-old OpenBSD flaw.
- Project Glasswing has been launched to utilize Mythos Preview for securing critical software and to prepare the industry for new cybersecurity practices.
- The model is also capable of reverse-engineering exploits on closed-source software and developing exploits for known (N-day) vulnerabilities.
- Researchers are limited in what they can disclose, as over 99% of the vulnerabilities found by Mythos Preview remain unpatched, necessitating coordinated defensive action across the industry.
Anthropic has unveiled Claude Mythos Preview, a new language model demonstrating unprecedented capabilities in cybersecurity, including the ability to identify and exploit zero-day vulnerabilities across major operating systems and web browsers. This development has prompted the launch of Project Glasswing, an initiative to leverage Mythos Preview for securing critical software and preparing the industry for advanced cyber threats. The model's proficiency extends to reverse-engineering exploits and turning N-day vulnerabilities into functional exploits, marking a significant leap in AI's role in cybersecurity.
