A Bloomberg developer claimed to have cracked an Enigma-encrypted Wehrmacht message from 1941 using OpenAI's GPT-6 Astra. The 82-character radio message, sent by a German soldier requesting marching orders, had remained unsolved for 83 years. The message was part of an archive of intercepted German army messages, listed as unsolved until now.
According to a detailed case study, Carter Leffen, a product development coach at Bloomberg LP, used the 'GPT-6 Astra Extra High' variant to decrypt the message. The AI model worked for about ten hours, handling tasks such as searching historical archives, building an Enigma simulator, and testing competing keys. The process involved cross-checking results and running parallel experiments.
The breakthrough came from an already-decrypted message sent on the same day, which contained the town name 'Rosenow' twice. Leffen and his AI agents used this as a search term, which helped crack the puzzle.
A known weakness of the Enigma machine, which never encrypts a letter as itself, also aided the decryption process. This allowed the team to rule out many positions and identify the correct settings.
The decrypted message included coherent German text, such as 'Sofort Funkantwort' and 'Angabe des Marschweges'. It also contained typos that Leffen believes are authentic, as a fabricated result would likely be error-free. The decrypted text was confirmed by a message header preserved separately in the archive.
GPT-6 Astra and several specialized AI agents served as Leffen's tools throughout the process. While the AI handled many tasks, Leffen emphasized that building the website to describe the problem and solution required more effort than the actual decryption. All code, search data, and a working 3D Enigma simulator are available for download on the project page.
Source: thedecoder
