OpenAI's Astra and Anthropic's Claude Opus Crack Long-Standing Enigma Messages
Two cryptanalysts have successfully decoded two long-unsolved messages from the WWII-era Enigma machine using large language models (LLMs) developed by OpenAI and Anthropic. Carter Leffen used OpenAI’s Astra to decode a message that had baffled researchers since 2005, while Jack Willis employed Anthropic's Claude Opus 5 for another unsolved cipher. Both AI systems demonstrated advanced research capabilities, including archival analysis, simulator construction, and pattern recognition far beyond typical human researcher timelines. Frode Weirerud, a cryptology expert who maintains the Crypto Cellar database of Enigma messages, validated both solutions and expressed astonishment at how quickly Astra completed what would have taken weeks for humans. The AI models also referenced data from sources outside his collection, raising questions about their access to private or public archives. This breakthrough highlights a new frontier in AI capabilities—leveraging machine learning not just for pattern recognition but as active researchers and problem solvers across historical and technical domains.
Topics: AI Alan Turing cryptanalysis Enigma code OpenAI