Ciphers are a wartime staple, yet French Emperor Napoleon Bonaparte learned that the hard way when one of his officers was captured in the Iberian Peninsula with a letter containing sensitive military information. Napoleon couldn’t quite grasp the need for encryption until this incident, after which he promptly tasked his army with developing a code. An engineer using OpenAI’s new model, GPT-6 Astra, has decoded one such unsolved cipher, allowing us to read a 217-year-old letter for the first time.
Translating the Coded Message
Carter Church, an AI engineer at cybersecurity firm Sentinel One, gave the AI model one prompt and a low-resolution scan of the letter published in a French army journal in 1969. Six hours later, the model decoded the message that had remained undeciphered for over two centuries. While models often make outrageous claims, the solution in this case could be mechanically tested because an incorrect key could not produce a historically consistent, coherent output.
Eugène, Napoleon’s stepson and the viceroy of Italy, sent the letter on the Emperor’s behalf to Marshal Auguste de Marmont, stationed far from other French troops on the eastern shore of the Adriatic Sea. The letter carried information about the position of every other French and allied army across the continent. To avoid revealing military secrets should the letter fall into the enemy’s hands, the message was written in code.
Church found the letter after searching for a suitable code to break, which also demonstrates GPT-6 Astra’s ability. “The verifiability of solutions is actually the most difficult part, far more difficult than generating potential solutions,” Church said. “You simply cannot automate accountability.” The letter had 155 different signs— 1,300 signs in total. Common letters have several interchangeable symbols, also called a homophonic code.
The model transcribed the text to find a partial table by cryptology historian Daniel Tant— previously unlinked to this letter. This data documented what 33 of the symbols meant, already accounting for about a third of the letter. Astra made the connection on its own, identified the partial cipher table as an example, and applied it to this cipher. The method worked, and GPT-6 translated the entire message by decoding what each other symbol indicated.
I used GPT-6 Astra to break an unsolved cipher to one of Napoleon's generals that had gone unread for 217 years. What makes this impressive isn't actually the codebreaking, but that Astra completed the entire multi-modal workflow in ~6 hours from a single image and goal. 1/6🧵 https://t.co/vyiGKOsrHq
— Carter Church (@CarterWChurch) September 30, 2026
Why is This a Big Break?
To understand why deciphering coded messages matters, we have to go back to World War II. Great Britain was on the brink of starvation as German U-boats ambushed merchant ships in the Atlantic Ocean. They communicated using encrypted messages delivered by Enigma machines. At this time, Alan Turing, a British mathematician, and Gordon Welchman developed the Bombe, a machine that decoded Enigma codes and disrupted them.
The highest accolade in computer science is named after Turing. He was also one of the founding fathers of AI, who asked whether machines could imitate human intelligence. GPT-6 Astra just gave a quite clear answer.
Click on Carter Church’s website for more information.
Contributors
Copyright @smorescience. All rights reserved. Do not copy, cite, publish, or distribute this content without permission.
