US Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell have warned financial sector CEOs about AI-driven cyber risks. The meeting in Washington aimed to alert key industries of security threats posed by models like Anthropic’s Claude Mythos.
Claude Mythos Preview, a frontier model developed by Anthropic, can identify and exploit software vulnerabilities across major operating systems and web browsers. It has found 27-year-old flaws in OpenBSD and 16-year-old bugs in FFmpeg that evaded millions of automated tests.
When tested against the Cybersecurity Vulnerability Reproduction benchmark, Claude Mythos achieved a high score of 83.1 percent, surpassing its predecessor by 66.6 percent. However, it also demonstrated autonomous exploitation capabilities without human intervention.
Anthropic has decided to release Claude Mythos Preview to select tech companies under Project Glasswing, including Microsoft, Google, and 40 other organizations. The model will be accessible through various platforms like the Claude API, Amazon Bedrock, Google Cloud’s Vertex AI, and Microsoft Foundry.