Anthropic announces Claude Mythos for cybersecurity research

Anthropic unveiled Claude Mythos Preview, an AI that can autonomously identify zero‑day vulnerabilities and develop related exploits.

Mythos Preview uncovered critical issues in OpenBSD, FFmpeg, and the Linux kernel and found thousands of high‑severity vulnerabilities across major OSes and web browsers.

All discovered vulnerabilities were reported and patched.

Access is restricted under Project Glasswing to a select group of partners — Cisco, AWS, Microsoft, CrowdStrike, JPMorgan Chase, Google, and other major tech and security companies — while open source maintainers can apply via Claude for Open Source.

Access is provided through Claude API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry; partners and open source communities are offered $100 million in model usage credits; a general public release is not planned.