Claude discovers cryptographic weaknesses in HAWK and AES

Anthropic researchers used Claude Mythos to discover mathematical flaws in the HAWK algorithm and a weakened version of AES, according to a blog post by Simon Willison. The findings have no practical impact on current systems. The model worked for 60 hours at an estimated $100,000 API cost, with key human interventions being prompts to not give up and to find something worth publishing. The work also led to a new evaluation benchmark, CryptanalysisBench, developed in partnership with ETH Zurich, Tel Aviv University, and University of Haifa.