Home / Technology / AI 'Mythos' Finds Critical Software Flaws
AI 'Mythos' Finds Critical Software Flaws
8 Apr
Summary
- Claude Mythos AI autonomously discovered thousands of zero-day vulnerabilities.
- It found a 27-year-old flaw in secure OpenBSD operating systems.
- The AI model is developed for defensive cybersecurity purposes.
Anthropic has unveiled Claude Mythos, a powerful frontier AI model developed under Project Glasswing. This AI is engineered for defensive cybersecurity, having demonstrated an extraordinary ability to autonomously discover thousands of previously unknown zero-day vulnerabilities in recent weeks. Its findings include a 27-year-old flaw in the highly secure OpenBSD operating system, critical issues within the widely used FFmpeg framework, and several Linux kernel vulnerabilities enabling complete machine control.
Recognizing the dual-use potential of such advanced AI, Anthropic is intentionally not releasing Claude Mythos to the public. Instead, the company is collaborating with 11 partners, including major tech firms and security companies, through Project Glasswing. The initiative aims to develop robust cybersecurity safeguards to detect and block dangerous outputs, paving the way for safe, large-scale deployment of Mythos-class models for defensive purposes.
