- Published on
OpenAI GPT-6 Astra Cracks Napoleon’s 1809 Coded Letter, Revealing Military Deployments From 217 Years Ago
- Authors
- Name
- DeepHorus
An AI engineer, using OpenAI's GPT-6 Astra model, successfully deciphered an encrypted military letter written by Napoleon Bonaparte in 1809. Because its codebook had been lost, the letter had remained incompletely deciphered for the past 217 years. The results were recognized by Satoshi Tomokiyo, the maintainer of Cryptiana, a historical cryptography research database.
The decipherment was carried out by Carter Church, an AI engineer and employee of SentinelOne. He said that he used the GPT-6 Astra model, starting from a scanned image of the historical document, to perform optical character recognition and cryptanalysis. The entire process took approximately six hours. This level of efficiency far exceeded that of traditional manual decipherment.
The letter was written on March 16, 1809, when Austria was preparing to go to war with France. Napoleon wrote to his stepson, Eugène de Beauharnais, the Viceroy of Italy, asking him to convey military information to the French general Auguste de Marmont through an encrypted letter. The surviving document begins with a line of ordinary French, followed by 24 lines of encrypted text composed of numbers, letters, and invented symbols.
The letter contains approximately 1,300 cipher units and 155 different symbols. Some of the symbols represent individual letters, while others correspond to complete French words. Researchers had previously identified the meanings of 33 of the symbols, but because the complete codebook had been lost, the remaining portions had continued to resist decipherment.
GPT-6 Astra first analyzed the scanned image, divided the document into multiple regions, and recognized the symbols within them. It then used methods including simulated annealing algorithms to conduct extensive testing of the possible letter and word combinations corresponding to the different symbols. During the analysis, the AI referred to the characteristics of historical French-language texts by Dumas père, Victor Hugo, and Marmont himself, in order to assess the plausibility of the different decipherment results.
The deciphered content revealed information about Napoleon's military deployments on the eve of the outbreak of the Austrian war in 1809. The letter listed the strength of French and allied forces in Bavaria, Poland, Saxony, and Italy, as well as the movements of the Russian army advancing toward Austria, supplementing details that had been missing from earlier historical records.
To verify that the model was not simply generating the answer by relying on existing historical materials, Church reran the analysis process after removing historical texts related to Napoleon and Marmont from the reference materials. The model still produced the same interpretation. Cryptiana subsequently marked the letter as having a deciphered code.
Church said that, in theory, this type of decipherment could also be performed manually, but it would normally require an enormous amount of time and specialized knowledge. The distinguishing feature of this AI system was that it could simultaneously perform tasks such as historical-document recognition, symbol classification, cryptological reasoning, and text reconstruction—tasks that had previously been handled separately by experts in different fields—and it did so in only six hours.
