GPT-6 Astra cracks 20-year-old unsolved WWII Enigma message
Original: OpenAI GPT–6 Astra breaks Enigma message that has resisted solution since 2005
Why This Matters
Demonstrates an AI independently conducting end-to-end cryptanalysis on a real unsolved historical problem without human guidance.
On September 15, 2026, researcher Carter Leffer reported that OpenAI's GPT-6 Astra independently broke MVUEH, a German Army Enigma message from July 10, 1941 that had resisted cryptanalysis since 2005. The AI autonomously wrote the necessary software, selected its own crib, and recovered the correct key and plaintext.
The MVUEH message was transmitted by a German radio station with the tactical callsign 2ny and logged as Nr. 172 by the SS-Totenkopf Quartiermeister Ib on July 10, 1941. Since its publication in 2005, all human attempts to break it had failed. A related message from the same day, Nr. 173 (SIPVX), was solved in 2017 by Alex Shovkoplyas, but the key he recovered did not unlock MVUEH.
Leffer gave GPT-6 Astra a single directive: try to break any of the unsolved Enigma messages listed on the Crypto Cellar Research website. The model independently identified MVUEH as the most tractable target, hypothesized that its plaintext might be related to SIPVX, wrote Python and C++ code to build an Enigma simulator and a Bombe, and settled on the repeated place name ROSENOW ROSENOW as a crib. That approach ultimately recovered the correct key.
The key turned out to be entirely different from the daily Enigma key used by all other messages from that date — even the wheel order differs (253 vs. 512). Analysis of the break also revealed transcription errors in the original ciphertext and an unusually rare left-hand wheel turnover at the 72nd letter, both of which likely stymied earlier solvers. The decrypted plaintext is nearly identical to SIPVX, differing only by 12 letters due to a misspelling of 'Bitte' and a repeated sender signature. Researchers say they are still reviewing the full GPT-6 Astra session logs to reconstruct its exact reasoning process.