Anthropic’s Mythos AI Finds Weaknesses in Cryptographic Algorithms

Anthropic says Claude Mythos Preview identified mathematical weaknesses in cryptographic algorithms tied to digital security. The model developed an improved attack on HAWK, a post-quantum signature scheme, and a new attack on a reduced version of AES-128. Anthropic says neither finding affects systems currently in use, but the work points to how AI could pressure-test assumptions behind internet security.

HAWK is described as one of the remaining schemes in the third round of NIST’s competition for additional post-quantum signatures. According to Anthropic, human experts had reviewed HAWK for more than two years before Mythos Preview found an improved attack in 60 hours. The report says the attack exploited a previously undetected symmetry in the mathematical lattice that HAWK relies on, with API costs totaling about $100,000.

Anthropic says Mythos also found a new attack on a reduced version of AES-128, not the full system used in practice. The attack applies to a modified AES version using 7 of the full scheme’s 10 rounds. The model developed a fingerprinting method Anthropic calls “Möbius Bridge,” which the article says improves on previously known attacks by a factor of 200 to 800.

The key takeaway is not that today’s deployed encryption is broken, but that AI systems may become powerful tools for cryptanalysis. Anthropic says Mythos worked largely on its own in a multi-agent setup, while human researchers mostly handled project management, prompting, and verification. The company shared findings with the U.S. government and industry partners, coordinated disclosure of the HAWK weakness, and continues to keep Mythos Preview unavailable to the public.