Anthropic has unveiled Claude Mythos Preview, an advanced AI model designed for cybersecurity, and launched Project Glasswing. The initiative grants select major technology and infrastructure organizations early access to the system for defensive security tasks.
Mysterious, the AI model has already identified thousands of high-severity vulnerabilities across major operating systems and web browsers. Anthropic aims to equip defenders with cutting-edge cyber capabilities before adversaries can develop similar tools.
Project Glasswing includes prominent partners like AWS, Apple, Google, Microsoft, and JPMorgan Chase. Anthropic is committing significant resources, including up to $100 million in usage credits and $4 million in donations to open-source security groups.
The Mythos model reportedly detects vulnerabilities and develops exploits with minimal human intervention, citing examples like a 27-year-old OpenBSD flaw and chained Linux kernel vulnerabilities. Anthropic states these identified bugs have been patched.
This announcement follows a security lapse where Anthropic accidentally exposed Claude's source code files due to a packaging error. The incident revealed nearly 2,000 files and over 500,000 lines of code, with a subsequent takedown effort affecting numerous GitHub repositories.
Anthropic is positioning Mythos as a highly capable and potentially dangerous model, opting for controlled release. The company plans to develop robust safeguards before a broader rollout, emphasizing safe deployment at scale.
An internal document described Mythos as Anthropic's most capable model yet, marking a significant advancement in reasoning, coding, and cybersecurity. The company has also engaged with U.S. government officials regarding the model's cyber implications.