Anthropic is testing the Mythos AI model for cybersecurity

Anthropic previewed Claude Mythos, which the company calls a frontier general-purpose model.

The preview is part of Project Glasswing, a defensive security initiative with about 12 core partners—including Amazon, Apple, Microsoft and the Linux Foundation—and ~40 additional organizations getting preview access.

In tests, Mythos reportedly identified thousands of vulnerabilities, including many zero-day flaws; many dated back 10–20 years and a large portion remain unpatched.

Claude Mythos was not specifically trained for cybersecurity, but improvements in code understanding and reasoning enabled these results.

Anthropic acknowledged exploitation risks, opted for a controlled rollout to a small group, and a leaked internal document (due to human error) plus a recent source-code exposure incident influenced caution.