Claude Mythos Preview finds 6,202 potential vulnerabilities estimated to be severity “High” or above, and an initial report is released on the high-performance AI now being adopted by Japanese banks

Anthropic’s initial Project Glasswing report says Claude Mythos Preview found more than 10,000 serious vulnerability candidates across critical software.

The company cited 23,019 candidates in over 1,000 open-source projects, including an estimated 6,202 high- or critical-severity cases.

The findings underscore how AI can speed up vulnerability discovery and validation, but also raise concerns about faster exploitation.

Japan’s finance ministry plans a public-private group to assess the model’s cyber risks, and Reuters says major Japanese banks may soon get access.