Anthropic disclosed that its Claude Mythos Preview model autonomously discovered previously unknown mathematical weaknesses in two cryptographic algorithms: HAWK, a post-quantum digital signature candidate, and a reduced-round version of AES, the encryption standard securing most of the internet.
Against HAWK, Mythos found a previously unnoticed symmetry in its lattice structure, letting researchers improve the best-known key-recovery attack and effectively cut the scheme's key strength in half — a result that took just 60 hours, despite HAWK having survived two years of expert human review. Against 7-round AES (the full cipher uses 10 rounds), Mythos devised a new fingerprinting technique Anthropic calls "Möbius Bridge," making the best-known attack 200 to 800 times faster by eliminating one of the guesses an attacker must make. Each result cost roughly $100,000 in API usage to produce.
Anthropic emphasized that neither finding threatens deployed systems: HAWK is only a standardization candidate, not yet in production, and the AES attack targets a deliberately weakened research variant, not the full 10-round cipher used in practice. The company says the work shows AI can now contribute original discoveries in theoretical cryptanalysis, a field long dominated by human mathematicians.