Enigma message unread since 2005 broken with help from GPT‑6 Astra
Cryptologic historian Frode Weierud says a German Army signal from July 1941 gave up its key after the model wrote its own Bombe in two days.
A German Army Enigma message that had resisted solution since 2005 has been read, according to Frode Weierud of Crypto Cellar Research, who says the cryptanalysis was carried out by OpenAI's GPT-6 Astra under the direction of Carter Leffer. Weierud's account, updated on 19 September, dates the break to 15 September.
The signal, known by its indicator MVUEH, was sent on 10 July 1941. Weierud says the model wrote its own Enigma simulation and Bombe software in Python and C++, then ran the attack using the repeated place name ROSENOW ROSENOW as a crib. It picked message number 172 as the most promising target on its own, he says, and suspected a link to a related message number 173.
The recovered key used wheel order 253, which Weierud notes differs from the other keys known for 10 July. He says the ciphertext also carried transcription errors, and that the left-hand wheel turned over at the 72nd letter, a rare event that complicates a crib attack.
Weierud, who has published Enigma material for decades, writes that what the model achieved in two days "would take a human researcher weeks or even months". That judgement is his own, and the claim rests on his account of the work. Neither OpenAI nor any other cryptanalyst has published an independent reconstruction of the break.
The page reached the front page of Hacker News on Tuesday with 696 points and 408 comments. Messages left unbroken after the wartime archives were opened are a small and well-worked set, and the ones that remain are usually corrupt, short, or both.