OpenAI GPT–6 Astra breaks Enigma message that has resisted solution since 2005

On 15 September 2026 Carter Leffer reported that OpenAI GPT-6 Astra had broken a long-unsolved German Army Enigma ciphertext, MVUEH (message no. 172) from 10 July 1941, which had resisted solution since 2005. MVUEH was sent by callsign 2ny to the SS‑Totenkopf Ib station and was logged at 17:30. A related message, SIPVX (no. 173), had been broken in 2017 by Alex Shovkoplyas, but its recovered key differed from the daily key and did not open MVUEH. The MVUEH key proved completely different (wheel order 253 versus 512) yet the plaintext is almost identical to SIPVX’s; MVUEH is 82 letters vs. SIPVX’s 94, the discrepancy explained by an encipherment error turning Bitte into Btte and a repeated sender signature, Waschbusch. The break exposed transcription errors in the published ciphertext and showed the Enigma’s left-hand wheel turned over at the 72nd letter, a rare turnover that complicates decryption.

GPT-6 Astra selected MVUEH autonomously, inferred a link to SIPVX, and used the repeated place name ROSENOW ROSENOW as a crib. It wrote Python and C++ code for an Enigma simulator and an Enigma Bombe and carried out a full cryptanalytic attack that produced the correct key and plaintext. Logs also show the model traced material to Bundesarchiv references RS 3-3/20a and RS 3-3/63b and a private collection, though it’s unclear whether those scans were actually accessed; researchers are still analysing the AI’s process. The author notes the AI achieved in two days what would likely take human researchers weeks or months and describes the result as remarkable.