Discovering cryptographic weaknesses with Claude
Anthropic researchers used Claude Mythos to discover cryptographic weaknesses in both HAWK and a reduced-round version of AES, though the paper notes "neither of these results has a practical impact on today’s computer systems." The researchers shared the prompts used, including misspellings, to encourage the model to persist: "the models tend to think it is impossible to solve so they don't try they need a good amount of prompting." Mythos Preview worked for 60 hours total at an estimated API cost of ~$100,000, with main human interventions being to urge it not to give up and "find something that worth publishing." The work also produced the paper "CryptanalysisBench: Can LLMs do Cryptanalysis?" describing a new evaluation created in partnership with ETH Zurich, Tel Aviv University, and University of Haifa.
LLMs can now assist in cryptographic research, potentially automating discovery of mathematical weaknesses.