Claude Mythos Preview Finds Attack Halving HAWK Key Strength
The results show an AI model improving on expert-reviewed cryptographic attacks, which Anthropic says points to large language models taking on part of the cryptography review role.
Reporting from 1 source: GIGAZINE.
Anthropic's Claude Mythos Preview found flaws in two encryption algorithms. It improved an attack on the HAWK digital signature scheme, effectively halving its key strength, and sped up cryptanalysis of a weakened AES variant. Neither result affects current systems.
Anthropic used Claude Mythos Preview to scrutinize multiple encryption algorithms and found flaws in two. The first is an attack on HAWK, a digital signature scheme that remained a candidate in NIST's 2022 solicitation for quantum-resistant cryptography. HAWK had passed a two-year expert review, but Claude Mythos Preview improved existing attack methods in 60 hours of work, effectively reducing the key strength by half. The estimated API cost for the process was about $100,000.
The second flaw concerns AES, adopted by NIST in 2001. Claude Mythos Preview improved the method for breaking a weakened version of AES built for verification, increasing the attack speed from 200 times to 800 times. Anthropic noted that neither result affects current systems, and that HAWK is not deployed while the AES result does not break the full cipher.
Synthesized by Yomimono from the 1 cited source below, including Japanese-language reporting where cited, then editorially reviewed before publishing.