News
AI · Anthropic · Cybersecurity · Frontier Models
Anthropic announced a preview of Claude Mythos, a new frontier AI model demonstrating remarkably strong cybersecurity capabilities, including the detection and exploitation of thousands of vulnerabilities, with Amazon, Apple, Microsoft, and the Linux Foundation among twelve core partners participating in its controlled deployment.
Claude Mythos is a general-purpose language model, not specifically trained for cybersecurity, but its enhanced code understanding and reasoning enable new applications. In tests, the model identified thousands of vulnerabilities, including many zero-day bugs reportedly present in software for ten to twenty years, with a large portion remaining unpatched.
This aligns with previously leaked internal descriptions that positioned Mythos as Anthropic's most advanced model, surpassing existing Opus models. Anthropic acknowledges the inherent risks of such powerful models, opting for a limited rollout to prevent misuse, while suggesting the underlying AI progress is difficult to halt.
The timing is notable as Anthropic recently experienced its own security issues, temporarily exposing thousands of source code files due to a software release error.
