OpenAI's Astra and Anthropic's Claude crack two unsolved Enigma codes

Alan Turing is best remembered for his namesake test of machine intelligence, but the more consequential test he faced in his own lifetime was cracking the Enigma code Nazi Germany used during World War II. Turing’s team built an early computer, the Bombe, that let the United Kingdom decode Enigma messages during the war, but a handful of archival messages have remained unbroken since, typically due to mistranscription or errors by the original encoders. Now, two independent cryptanalysts say they’ve used large language models from OpenAI and Anthropic to crack two of those long-unsolved messages.

Developer Carter Leffen told OpenAI’s newest model, Astra, to search a database of Enigma messages for one that remained unbroken and decode it. Astra did its own archival research, found context clues, built a simulator of the Enigma machine, and recovered the plaintext of a message that had baffled researchers since 2005. Leffen then used Astra to build an interactive website explaining the whole problem.

Frode Weierud, a retired electrical engineer with a lifelong interest in cryptology who maintains the Crypto Cellar website — a repository of cryptology resources, records, and a database of Enigma messages — validated Leffen’s solution last week, saying it left him in “awe.” Weierud noted that Astra’s logs referenced archived messages from a “private collection” that isn’t hosted on his own site; he isn’t sure whether the model actually accessed them, and speculates they may have been shared by another researcher online or that Astra reached the German government’s public archives directly. “What it has achieved in two days would take a human researcher weeks or even months,” Weierud wrote, adding that he personally spent several weeks researching the same Bundesarchiv files the model cited.

On September 21, cybersecurity executive and cryptanalyst Jack Willis told Weierud he had used Anthropic’s Claude Opus 5 to break a separate unsolved message, providing the model significantly more guidance than Leffen gave Astra. Claude ultimately used the known signature of a particular officer’s name to crack the code.

Weierud says just seven Enigma messages remain unbroken, along with one whose plaintext is already known but whose underlying code has not been recovered.

Why it matters

If general-purpose LLMs can now do archival cryptanalysis that used to take human researchers months, historians may be able to recover the small remaining pool of undeciphered wartime records — and the same research-and-pattern-matching approach could extend to other unsolved archival puzzles beyond Enigma.