AI password attacks improved... Zcash protected transactions use a different proof method

Anthropic said its Claude Mythos Preview model improved existing attacks on HAWK, a quantum-resistant signature candidate, and on reduced-round versions of AES, though neither result breaks a deployed system. HAWK has not yet been deployed in production, and the AES work targeted only cut-down round counts rather than the full algorithm, so live systems remain unaffected. Zcash’s (ZEC) Orchard shielded transactions rely on different proving cryptography than the primitives examined in the study, insulating them from the findings. The results matter as a stress test of post-quantum signature candidates ahead of eventual migration, and as a sign that AI models can sharpen cryptanalysis rather than only speed up code. Watch for follow-up work on full-round AES and for HAWK’s progress toward standardization and deployment.