OpenAI's Astra and Anthropic's Claude Opus 5 crack two long-unsolved Enigma messages

Two large language models — OpenAI's Astra (GPT-6) and Anthropic's Claude Opus 5 — have decrypted a pair of Nazi Enigma messages that had resisted researchers for decades. The messages belonged to a small set of archival intercepts left unread, largely because of transcription errors or mistakes by the original wartime operators.

Developer Carter Leffen asked Astra to locate an undeciphered Enigma message in the historical corpus and break it. The model conducted independent archival research, identified contextual clues, built a working Enigma simulator and recovered the plaintext of a message that had baffled experts since 2005. Leffen then used the model to create an interactive site that walks through the problem and the solution step by step.

Frode Weirerud, a retired electrical engineer who runs the Crypto Cellar site and curates an Enigma message database, verified Leffen's solution and said it left him "in awe." What the model accomplished in two days would have taken a human researcher weeks or months, he wrote. Weirerud also noted that Astra's logs reference messages from a "private" collection not held in his archive; it is unclear whether the model accessed that collection directly or located the material in public German government archives.

On 21 September another cryptanalyst, Jack Willis, contacted Weirerud to report that he had used Claude Opus 5 to crack a different unsolved message. Willis supplied the model with substantially more guidance, and it exploited a known signature — the name of a specific officer — to break the cipher.

Weirerud says seven Enigma messages now remain undeciphered, plus one whose plaintext is known but whose cipher has not yet been reconstructed. At the current pace, that list may shrink further soon.