AI model discovers cryptographic weaknesses in HAWK and AES

Anthropic's Claude Mythos Preview AI model has identified significant cryptographic weaknesses, leading to an end-to-end key-recovery attack against HAWK-256 and a substantial speedup for an attack on seven-round AES-128, according to a recent report by The Hacker News.

The HAWK attack exploits a previously unknown symmetry in the lattice structure of the signature scheme, enabling a key recovery in approximately three hours and 42 minutes on a 96-core server. This attack specifically targets the HAWK-256 challenge parameter and does not affect production systems, and HAWK remains a candidate in the NIST post-quantum standardization process.

The AES-128 attack, targeting seven of its 10 rounds, achieves a 200- to 800-fold speedup by removing a guessing step from an existing meet-in-the-middle attack. However, this attack still requires an impractical number of chosen plaintexts and does not necessitate changes to production software.

Anthropic emphasized that neither finding impacts current security protocols, with the HAWK attack not extending to other NIST signature candidates or lattice cryptography generally. The research, largely conducted by the AI with human oversight for direction and verification, highlights the growing capability of AI in cryptanalysis.

Source: The Hacker News