AI has decoded a Napoleon-era secret letter that had remained undeciphered by researchers for 217 years, revealing troop positions ahead of war with Austria in 1809. It was not written by Napoleon himself, but sent from the headquarters of his stepson, Eugène de Beauharnais, to General Auguste de Marmont on Napoleon’s orders.
A code hidden in 1,300 symbols
The letter appeared in a French military journal in 1969, but carried the wrong date. After one line of ordinary French, it contains 24 rows of coded numbers, letters and hand-drawn marks. The original key was lost.
Carter Church, an AI engineer at SentinelOne, used OpenAI’s GPT-6 Astra to analyse a low-resolution scan. It transcribed the marks, found a partial code table compiled by historian Daniel Tant, and worked out the remaining signs. Tant’s table covered 33 symbol values, about one-third of the text. The letter contains roughly 1,300 coded units and 155 distinct signs; the task took around six hours of model execution.
The code was “homophonic”: one letter could be represented by different symbols. Some marks also stood for whole words, making the puzzle harder to solve by counting letters alone.
What was hidden in the message?
The decoded briefing listed the reported positions and strength of several forces. It said 40,000 Bavarian troops were at Munich and Passau, 30,000 Polish troops were near the Vistula River, and French forces were stationed at places including Bayreuth and Ulm. It also claimed the army of Friuli had 80,000 men. These were figures reported in the letter, not independently verified headcounts.
The message warned that Austria would face pressure from several directions and urged Marmont not to fear small enemy groups. Marmont commanded French troops in Dalmatia, separated from other French forces by Austrian territory. Napoleon wanted him to understand the wider military situation as Austria prepared to fight France.
A breakthrough, but not a new war secret
The discovery does not reveal a completely unknown battle plan: historians already knew much of the background from other letters. Its value is that it makes the coded message readable and gives researchers another record to compare with historical evidence.
Church published the transcription, code key and a script so others can check the result. The historical-cipher website Cryptiana marked the puzzle solved after its curator reviewed the reading. Five symbols in the reconstruction remain uncertain, leaving room for further scrutiny. The case shows how AI can unlock records- but why evidence and human judgment still matter.
Discover AI Breakthroughs, Emerging Technologies and Future Career Trends!
1. Can AI Listen and Speak Simultaneously?
Discover how a new voice AI model handles listening and speaking at the same time.
2. What Makes GPT-6 Astra AI Different?
Explore the reported capabilities of GPT-6 Astra and what sets this AI model apart.
3. Why Is AI Model Distillation Controversial?
Uncover the debate surrounding AI model distillation involving DeepSeek and Moonshot.
4. Which AI Role Offers Better Opportunities?
Compare AI engineering and AI implementation specialist careers in India’s changing job market.
5. Are AI Skills Overtaking College Degrees?
Find out why observers are questioning the value of traditional degrees in India’s AI skills race.
