OpenAI GPT–6 Astra breaks Enigma message that has resisted solution since 2005 — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- GPT-6 Astra autonomously broke Enigma message MVUEH from 10 July 1941 — sent by radio station 2ny to the SS-Totenkopf Division's Ib (Quartiermeister) station — which had resisted all cryptanalytic attempts since 2005.
- Carter Leffer only directed the AI to try the unbroken Enigma messages on the Crypto Cellar Research web page; GPT-6 Astra independently selected message Nr. 172 as most promising and inferred its plaintext was related to Nr. 173 SIPVX.
- GPT-6 Astra built its own Python and C++ Enigma simulator and Bombe software, then used the repeated place name "ROSENOW ROSENOW" as a crib to recover the correct key.
- The recovered key differs entirely from the daily 10 July 1941 key used by other messages that day — wheel order 253 instead of 512 — which the researcher identifies as a key reason the cipher eluded earlier solvers.
- The 82-letter MVUEH plaintext is nearly identical to the 94-letter SIPVX message broken by Alex Shovkoplyas in 2017, differing only by a "Bitte"/"Btte" enciphering error and the sender's repeated Waschbusch signature.
- GPT-6 Astra also cited specific German Bundesarchiv file references (RS 3–3/20a and RS 3–3/63b) that are not on the Crypto Cellar Research website; the human researcher spent weeks researching these files himself and calls the AI's behavior "like a very professional cryptanalyst and archive researcher."
Why it matters: A message that defeated human cryptanalysts for 21 years — complicated by transcription errors in the original form and a rare left-hand wheel turnover at the 72nd letter — was broken in roughly two days by an AI that also independently produced its own simulator, Bombe, and archive file references. For the cryptographic history community, it marks the first documented case of an LLM performing the full codebreaker workflow, from crib selection to key recovery, without human guidance beyond a prompt.
Ask SkimNews
