Discovering cryptographic weaknesses with Claude

Anthropic researchers using the Claude Mythos model discovered improved cryptanalytic attacks: one that halves the effective key strength of the HAWK post-quantum signature candidate (requiring larger keys) and another that significantly speeds up known meet-in-the-middle attacks against a 7-round reduced AES-128 variant; both results are research advances with no immediate impact on deployed systems, and the team followed coordinated disclosure and published technical papers and reproducible code.