Discovering Cryptographic Weaknesses with Claude

Anthropic researchers using the company's Claude Mythos Preview model discovered improved attacks against two cryptographic algorithms, the company announced. The first attack significantly weakens HAWK, a digital signature scheme under consideration by the National Institute of Standards and Technology for post-quantum cryptography standards. The second identifies a new way to attack a round-reduced version of the Advanced Encryption Standard, the most widely used symmetric cipher. Anthropic said these are substantial research advances but do not currently affect any production systems. HAWK is only a candidate signature scheme and is not deployed, while the AES attack targets a reduced version and does not break the full cipher. The HAWK attack effectively cuts the scheme's key strength in half, improving the best-known attack in 60 hours of work despite the algorithm surviving two rounds of expert human review over two years. The AES attack improves the speed of previous best attacks by 200 to 800 times. Each result cost roughly $100,000 in API costs to develop. Anthropic said it followed responsible disclosure procedures, sharing the HAWK finding with the algorithm's authors in June and coordinating public disclosure on the NIST mailing list. The company also partnered with academics at ETH Zurich, Tel Aviv University, and University of Haifa to build CryptanalysisBench, a benchmark for evaluating large language models on cryptographic tasks.