Anthropic turned a candidate-scheme weakness into something other evaluators can test, which is the real shift here. The HAWK result is no longer just a claim about hidden structure; it now comes with reproducible code and papers, so the question becomes how much security margin HAWK-256 really has left in NIST review.
Anthropic says Claude Mythos Preview found a hidden symmetry in HAWK's lattice math and used it to cut the hard part of key recovery down to roughly half size for the HAWK-256 challenge parameter. The same work also sped up an attack on seven-round AES-128 by removing a 256-way guessing step, but Anthropic says neither result affects production systems and the AES attack still needs impractical data.
For PQC teams, the practical change is in evaluation, not incident response. A scheme can lose the efficiency edge that made it attractive before deployment, so candidate lists and parameter choices can move under you as new cryptanalysis lands.