Computer pioneer Alan Turing, best known for the eponymous Turing test, faced a more consequential challenge during his lifetime: breaking the Enigma cipher used by Nazi Germany in World War II. Although Turing and his team built an early machine called the Bombe that enabled Allied forces to read many Enigma messages during the war, a small number of archival Enigma messages have remained unsolved—often because of transcription errors or mistakes by the original encoders.
Two modern cryptanalysts now say they have used large language models to recover the plaintext of two of those long-unresolved messages. Developer Carter Leffen instructed OpenAI’s newest model, Astra, to search a database of Enigma messages for an unbroken entry and to decode it. According to Leffen, the model conducted its own archival research, found contextual clues, constructed a simulator of an Enigma machine, and ultimately recovered the plaintext of a message that had puzzled researchers since 2005. Leffen also used Astra to create an interactive website that explains the problem and the steps of the solution.
Frode Weirerud, a retired electrical engineer with a lifelong interest in cryptology who runs the website Crypto Cellar, maintains a variety of resources and a database of messages. Last week Weirerud validated Leffen’s solution and said he was left in “awe.”
Notably, the Astra model’s logs include references to archived messages in a “private collection” that are not hosted on Weirerud’s site. Weirerud said he is unsure whether the model actually accessed those materials; he speculates they may have been shared elsewhere online by a different researcher, or that the model may have accessed public holdings in the German government archives. In Weirerud’s words, “GPT–6 Astra is behaving like a very professional cryptanalyst and archive researcher. What it has achieved in two days would take a human researcher weeks or even months. Personally, I spent several weeks researching the Bundesarchiv files GPT–6 Astra refers to.”
Separately, on September 21, cryptanalyst Jack Willis contacted Weirerud to report that he had used Anthropic’s Claude Opus 5 model to break another previously unsolved message. Willis provided substantially more human guidance to Claude; the model ultimately used a known officer signature pattern (the known signature of a particular officer’s name) to break that message.
According to Weirerud, seven Enigma messages remain unbroken, and there is one additional message where the plaintext is known but the code has not been reconstructed. Given the recent successes by AI-assisted researchers, some of these remaining messages may be solved in the near future.
