Anthropic's Claude Model Discovers Post-Quantum Cryptography Flaws, Reduces HAWK Key Attack Cost from 2^64 to 2^38

According to Anthropic on July 28, its Claude Mythos Preview model discovered previously unknown attacks on HAWK, a post-quantum digital signature scheme competing for U.S. federal standard status. The breakthrough reduced the cost of stealing HAWK's smallest key from 2^64 operations to 2^38—roughly 67 million times less computational work.

Claude also accelerated attacks on a 7-round version of AES by 200 to 800 times, surpassing the previous record set by cryptographers in 2013. Each discovery required approximately $100,000 in API usage and several hundred hours of Anthropic staff verification time.