OpenAI GPT-6 Astra Successfully Decodes 20-Year-Old Enigma Message

OpenAI's latest model GPT-6 Astra successfully decoded a German Army Enigma message (MVUEH) that had remained undeciphered since 2005. After researcher Carter Leffer instructed the model to assess the decipherability of messages from Crypto Cellar Research's list of undeciphered messages, GPT-6 Astra selected Nr. 172 (MVUEH) as the most promising target and performed the work.

Decoding Process and Technical Approach

GPT-6 Astra began its analysis by hypothesizing plaintext correlations with similar messages that had already been decoded (Nr. 173, SIPVX). It utilized the repeated place name 'ROSENOW ROSENOW' as a crib and, based on this, directly developed Python/C++ software for an Enigma simulator and a Bombe to complete the decryption. The message was known to be highly difficult to decode because it used a different key and wheel order on the same day (253 vs 512), and a rare turnover phenomenon occurred at the 72nd character. Additionally, there were complex variables such as transcription errors in the original text and typos (e.g., 'Bitte' written as 'Btte').

Research Efficiency and Archive Access

GPT-6 Astra completed a task that would take human researchers weeks or months in just 2 days. Log analysis suggests that the model discovered and referenced notes from the digitized collections of the German Federal Archives (Bundesarchiv). While the file numbers referenced by the model (such as RS 3–3/20a) were confirmed to be accurate, the image for the original reception number 172 has not yet been located. Researcher Frode Weierud evaluated GPT-6 Astra's behavior as being very similar to that of professional cryptanalysts and archive researchers, acknowledging this AI decryption as an astonishing achievement. The update is as of September 19, 2026.