GPT-6 Astra has decoded a 1941 German military message that remained unsolved for decades, showcasing AI's growing capabilities in cryptography.
In a remarkable achievement, GPT-6 Astra has successfully decoded a German military radio message from July 10, 1941, which had remained unsolved in the CryptoCellar archive since 2005. This feat marks a significant milestone in the use of artificial intelligence for cryptographic challenges.
Just two days ago, GPT-6 Astra decrypted an unsolved German Army Enigma message from 1941, demonstrating its ability to autonomously search historical archives, identify contextual clues, and run cryptanalysis simulations.
— Carter Leffen (@carterleffen) September 17, 2026
The message, identified as MVUEH, consists of 82 encrypted characters. Attempts to decode it using known keys proved ineffective. The decoding process, led by Carter Leffen, took place over September 14 and 15 and involved Astra working alongside several other AI agents to analyze archival materials, compare various interpretations of the encrypted symbols, and generate potential keys.
Leffen described Astra's effort as taking approximately 10 hours, although the parallel nature of the tasks means it is difficult to distinguish between contributions from the AI and human researchers.
A Key Insight
A crucial lead originated from another decrypted message, SIPVX, from the same date, which mentioned the name "Rosenow" twice. This clue guided researchers to use this repeated sequence as a "crib" to assist in deciphering the MVUEH message.
A series of challenges arose due to the condition of the historical documents, with researchers noting instances where handwritten letters could be interpreted in multiple ways. Ultimately, Astra developed an Enigma simulator and cryptanalysis programs to sift through billions of possible combinations of settings to find a workable solution.
After extensive verification, the researchers established the rotor settings, ring settings, initial positions, and plugboard connections necessary to decode the message. Two replicable Enigma implementations confirmed the accuracy of the decoded content.
Message Details
The decoded radio message inquired about the sender's route, confirmed they were in a location referred to as Rosenow, and requested an urgent radio response. The exact location of Rosenow remains unclear, though the sender's signature was preliminarily identified as Waschbusch.
Frode Weyerud, an expert in Enigma decryption, affirmed the findings and highlighted that the key obtained differed significantly from previously known keys from the same date. He noted that earlier decoding attempts might have faltered due to transcription errors and complex rotor configurations.
The researchers clarified that this achievement does not indicate a breakthrough against the Enigma cipher itself, as the techniques for such cryptanalysis were already known during World War II. Instead, they restored the specific key and text of a previously unsolved message.
This development follows another recent AI accomplishment where OpenAI claimed to solve a major mathematical problem in fluid dynamics, demonstrating the growing capabilities of AI technologies in tackling complex problems.
