GPT-6 Astra broke an Enigma message stuck since 2005

OpenAI’s GPT-6 Astra recovered the key and plaintext of MVUEH, an 82-letter German Army Enigma message from 10 July 1941. It started from a single instruction and wrote its own codebreaking software along the way. Frode Weierud, who maintains the list of unbroken messages the model was pointed at, confirmed the break in a write-up on Crypto Cellar Research, last updated on 19 September.

Carter Leffer contacted Weierud on 15 September and asked him to validate the result. Leffer’s only direction to the model had been to see whether it could break any of the unbroken Enigma messages published on the Crypto Cellar site. Once Weierud saw the details, “it was immediately clear he had found the correct key and plaintext.” Every attempt to break the message since 2005 had failed. Tom’s Hardware says that was the year the message was shared with enthusiasts online.

Astra picked its own target. After going through the unbroken messages, it judged Nr. 172, MVUEH, the most promising. It also quickly suspected a link to Nr. 173, SIPVX, another message from the same day, which Alex Shovkoplyas broke on 9 June 2017. MVUEH came from a station with the tactical callsign 2ny and was logged at 17:30 by the radio station of the SS-Totenkopf Division’s quartermaster section, Ib. The hunch was right: the two plaintexts are almost identical. Per Hardware Busters, the decrypt asks for the route of march, gives the sender’s position as Rosenow, twice, requests an immediate radio reply and is signed Waschbusch. SIPVX runs to 94 letters against MVUEH’s 82. Weierud explains the gap: the operator enciphered “Bitte” as “Btte” in MVUEH, and SIPVX repeats the signature.

That repeated place name became the crib. A crib is a stretch of plaintext you are confident appears in the message. The Bombe was the machine Allied codebreakers used to test rotor settings against a crib and discard every setting that could not produce it. Astra wrote an Enigma simulator and a Bombe in Python and C++, and after “trying many different approaches” it ran a full search with ROSENOW ROSENOW. That search returned the correct key. The method is the same crib-and-bombe approach Bletchley Park used in the war.

The MVUEH key is completely different from the other two keys of 10 July 1941, down to the wheel order: 253, against 512 for both the daily key and the SIPVX key. Shovkoplyas’s key had already broken the pattern, with the daily wheel order but different plug connections and ring settings, and it still did not open MVUEH. The break also showed that the ciphertext transcribed from the original message form contained several errors. And the left-hand rotor turns over at the 72nd letter, a rare event known to complicate a break. “These factors may have prevented an earlier break,” Weierud writes. Hardware Busters adds that compute was not the obstacle, since a three-rotor Army Enigma is well within reach of a modern PC once you have a decent crib.

Weierud is still analysing Astra’s logs to see exactly how it carried out the break, “and we are discovering amazing details.” One of them concerns the archive. In July 2026 he had posted a note on the 1941 message list about radio-message collections found in the German Bundesarchiv, among them messages from the division’s logistics command, the Nachschubführer. Astra appears to have found that note, because one of its logs reads:

Original-source access lead. The completed evidence pass traced the received corpus to a private collection and found concrete Bundesarchiv radio-message volumes, including RS 3–3/20a and RS 3–3/63b. Catalogue records have been inspected; the original received no.172 image remains unlocated, and the available archive viewer has not yet exposed the relevant scans in this environment. No correspondence has been sent.

Weierud confirms that the file references RS 3-3/20a and RS 3-3/63b are correct, and they are not available on the Crypto Cellar Research website. It is unclear what the “private collection” is, and whether Astra got into the Bundesarchiv’s digitised collections or found the files somewhere else. “What it has achieved in two days would take a human researcher weeks or even months,” Weierud writes.

He himself spent several weeks researching the same Bundesarchiv files.