Anthropic Releases Claude Mythos: AI Model with Striking Cybersecurity Capabilities, Launches Project Glasswing
Anthropic has released Claude Mythos Preview, a new general-purpose language model that performs strongly across the board but is "strikingly capable" at computer security tasks. In response, Anthropic launched Project Glasswing to help secure critical software and prepare the industry for the new era of AI-powered cybersecurity.
Key Capabilities
The Mythos Preview model has demonstrated remarkable abilities:
- Zero-day discovery — Finding and exploiting previously undiscovered vulnerabilities in real open-source codebases
- Closed-source reverse engineering — Reverse-engineering exploits on closed-source software
- N-day weaponization — Turning known-but-unpatched vulnerabilities into working exploits
- Over 99% of vulnerabilities found remain unpatched — Anthropic is following coordinated disclosure
Project Glasswing
Anthropic's coordinated effort to reinforce global cyber defenses includes:
- Technical evaluation methodology for AI security capabilities
- Coordinated vulnerability disclosure process for AI-discovered bugs
- Industry preparation guidelines for the new AI-powered security landscape
Strategic Partnerships
Bloomberg reports that Apple and Amazon are already testing the more powerful Mythos AI model, signaling rapid enterprise adoption.
What This Means
Anthropic calls this a "watershed moment for security" — the model can find vulnerabilities that human auditors miss, but the same capabilities could be used by malicious actors. The dual-use nature of AI security tools has never been more apparent.
The model's system card (93 points on Hacker News) has generated significant discussion in the security community about both the promise and risks of AI-assisted vulnerability discovery.