• Sources: Anthropic research, anthropics/cryptography-research-demo, HN 49087091
  • Summary: Anthropic published research reporting that its Claude Mythos Preview model found weaknesses in existing cryptographic constructions. Two results are stated. The first halves the effective key strength of the HAWK signature scheme, with a published example of full key recovery against HAWK-256 falling from 2^64 to 2^38. The second speeds up an attack on 7-round AES by 200 to 800 times over the previous best meet-in-the-middle attack. Anthropic states that neither result has a practical impact on today's computer systems and that no production software has to change as a result: HAWK is a candidate signature scheme and is not deployed, and the AES result is against a reduced-round version rather than the full cipher. Anthropic says it consulted academics to confirm the validity of the findings, shared the HAWK attack with the scheme's authors in June, and coordinated disclosure to the public NIST mailing list at release. No third-party reproduction is published. A companion repository, anthropics/cryptography-research-demo, accompanies the write-up.
  • Why it matters: HAWK is a live third-round NIST post-quantum signature candidate that survived two years of expert review, and doubling its key size to restore the intended security level removes most of the reason to pick it.
  • Follow-up: Watch for a third-party reproduction of the HAWK result, and for how the NIST process treats the candidate after the mailing-list disclosure.

send feedback on this story