Anthropic's Mythos AI finds flaws in internet encryption standards

What happened

Anthropic's Claude Mythos Preview AI model discovered an improved attack on the post-quantum signature scheme HAWK in 60 hours for $100,000 in API costs, and found a new attack on a reduced version of the Advanced Encryption Standard (AES) in roughly 1 billion tokens for about $100,000. The HAWK attack exploits a previously undetected symmetry in the mathematical lattice underlying the scheme; the AES attack uses a new fingerprinting method called "Möbius Bridge" that improves on the best previously known attacks by a factor of 200 to 800.

Why it matters

AES is the world's most widely used symmetric encryption standard for digital data, and HAWK is a candidate in the U.S. National Institute of Standards and Technology (NIST) post-quantum standardization process. Although Anthropic says neither finding affects systems in use today—HAWK is not yet standardized and the AES attack applies to a modified 7-round version rather than the full 10-round scheme—the findings demonstrate that AI models could challenge core assumptions behind internet security and uncover vulnerabilities that human experts reviewing over two years did not detect.

What to watch

Anthropic shared its findings in advance with the U.S. government and industry partners and coordinated disclosure of the HAWK weakness with the scheme's authors. Mythos Preview remains unavailable to the public. Anthropic has also developed a benchmark called CryptanalysisBench with researchers from ETH Zurich, Tel Aviv University, and the University of Haifa to let others systematically evaluate the cryptanalytic abilities of language models.