Claude Model Finds Cryptographic Flaws in HAWK and AES
Anthropic disclosed two attacks its unreleased Claude Mythos Preview model helped researchers uncover: a key-recovery attack on the post-quantum signature scheme HAWK, and a related "Möbius bridge" attack on AES. Anthropic published full technical papers for both, including the model's chain-of-thought reasoning behind the AES attack, and built CryptanalysisBench with academics from ETH Zurich, Tel Aviv University and the University of Haifa to benchmark LLM cryptanalysis ability going forward.
Why It Matters
Publishing the model's reasoning trace behind an offensive-security result is an unusual level of transparency, and signals cryptanalysis is becoming a domain where LLMs contribute original findings, not just triage known ones.