Why it matters
Anthropic reports that Claude Mythos Preview helped produce an end-to-end HAWK-256 key-recovery attack and a projected 200- to 800-fold speedup for an attack on seven-round AES-128. Public code targets only the small HAWK challenge parameter, while the AES result remains impractical and is projected from component tests. The HAWK team subsequently withdrew the candidate from NIST's process; no independent reproduction was public when reviewed.
My takeaway: Treat this as evidence of research acceleration and verification cost, not a break of production AES or lattice cryptography. Capability evaluations should separate end-to-end reproductions from projected components, record human and model labor plus compute, require external reproduction, and measure downstream impact—here, withdrawal of a standards candidate.