Crypto
Home›Crypto›Regulation›AI model finds new attacks on post-quantum signature H…
AI model finds new attacks on post-quantum signature HAWK and AES
The work cut the cost to steal HAWK’s smallest key from 2^64 to 2^38 operations, and sped up an attack on a 7-round AES version by 200 to 800 times.
Anthropic’s unreleased Claude Mythos Preview model uncovered two previously unknown cryptographic attacks, including one against HAWK, a post-quantum digital signature scheme being developed for potential U.S. federal standardization. According to Decrypt, Claude found a symmetry in HAWK’s math that reduced the estimated work to recover the scheme’s smallest secret key from 2^64 operations to 2^38, a drop of roughly 67 million times. NIST placed HAWK into the third round of its post-quantum signature competition in May, where it is described as the last lattice-based candidate standing. The AI team said fixing the HAWK weakness would likely require roughly doubling HAWK’s key size, which, in its view, would reduce the scheme’s attractiveness as a quantum-resistant candidate by eroding its pitch of compact keys and fast signing. Anthropic also coordinated the findings with the algorithms’ authors and U.S. government and industry partners before publishing, and worked with NIST on the disclosure. Decrypt also reported that Claude sped up an attack on a 7-round version of AES by 200 to 800 times versus a 2013 record, with each result costing roughly $100,000 in API usage. The company said the AES effort was verified by Anthropic staff over several hundred hours, and that HAWK has not been deployed while Bitcoin currently runs on ECDSA, not HAWK, “for now.”
Latest closeBitcoin $63,547.54 ▼0.3%