Astra and Claude Opus 5 crack unsolved Enigma messages

Alan Turing is most famous for his test of machine intelligence, but the work he did cracking Nazi Germany's Enigma code during World War II may have been more consequential in his lifetime. His team built an early computer, the Bombe, to translate Enigma messages, yet a handful of archival messages stayed unbroken, usually because of mistranscription or errors by the original encoders. That leftover puzzle set is what two cryptanalysts have now begun to clear using large language models from OpenAI and Anthropic.

The first came from developer Carter Leffen, who simply told OpenAI's Astra to search a database for an unbroken message and decode it. The model did so after conducting its own archival research, finding context clues, and building a simulator of the Enigma machine. Retired electrical engineer Frode Weierud, who maintains the Crypto Cellar website and its records, validated the solution last week. He noted that the model's logs referenced archived messages in a private collection not hosted by him, and he remains unsure whether the model accessed them or whether they were shared elsewhere online or pulled from the German government's public archives. He wrote that Astra behaved like a professional cryptanalyst and archive researcher, achieving in two days what would take a human weeks or months.

On September 21, another cryptanalyst, Jack Willis, told Weierud he had used Anthropic's Claude Opus 5 to break a different unsolved message, though he gave the model significantly more guidance and it relied on a known officer's name signature. With only seven unbroken Enigma messages and one partially known one left, the pace of these solves suggests the archive may not stay unresolved for long. How quickly the remaining messages fall may depend on how much human guidance each model is given, and on whether the models' own research methods can be replicated and verified by the cryptology community.