AI Decodes 200-Year-Old Napoleonic Wartime Letter
The breakthrough demonstrates how large language models can resolve legacy ciphers in hours rather than decades.
Updated on Oct. 16, 2026 in Artificial Intelligence
Live Poll
Do you feel that AI solving long-standing historical puzzles adds value to our understanding of history?
Artificial intelligence has successfully decoded a 1809 letter from a Napoleonic general that remained unsolved for over 200 years. Researcher Carter Church utilized the GPT-6 Astra model to translate the encrypted wartime troop positions in just six hours.
Why it matters
The deployment of AI for cryptanalysis and manuscript transcription demonstrates a shift away from labor-intensive manual methods. This capability enables the rapid processing of historical datasets that were previously considered too fragmented or complex for traditional scholarship.
The decryption process required six hours of compute time using the GPT-6 Astra model. While performance metrics for historic cipher-breaking are not standardized, this result follows a trend where AI systems have solved the majority of ciphers listed on the Cryptiana website since mid-September 2026.
The players
Carter Church
An independent researcher who successfully applied large language models to legacy cryptanalysis.
Satoshi Tomokiyo
A researcher based in Tokyo who maintains the Cryptiana website tracking historical unsolved ciphers.
OpenAI
The developer of the GPT-6 Astra model used to translate historical documents.
The details
The decryption relied on the large-scale pattern recognition capabilities of the GPT-6 Astra model to identify linguistic structures within the encrypted text. Unlike traditional cryptanalysis, which requires manual hypothesis testing of potential ciphers, this approach applies high-dimensional probabilistic modeling to map the code against historical French and allied military terminology. Researchers are increasingly applying similar character recognition systems, such as the one used for the e-NDP project, which achieves 88 to 94 percent accuracy on medieval administrative Latin script.
Timeline
79 AD: The Vesuvius eruption carbonized the Herculaneum Papyri.
1809: Napoleon ordered the writing of the coded wartime letter.
June 2026: Researchers performed the first virtual unrolling of a Herculaneum scroll.
Mid-September 2026: AI solutions for the Cryptiana cipher list accelerated.
October 2026: Satoshi Tomokiyo was photographed in Tokyo.
The Tech Race
This achievement follows the recent success of the e-NDP project at Notre Dame Cathedral, which currently utilizes AI to transcribe 14,600 pages of Latin archives. Both efforts sit at the leading edge of a broader field that is currently applying deep learning to recover data from thousands of years of human history.
These results suggest that professional historians and hobbyists will soon have access to automated tools for transcribing and decoding complex legacy documents. For researchers, this means that archives previously locked behind high labor costs are becoming accessible for real-time analysis.
The takeaway
AI is rapidly closing the gap on centuries-old cryptology problems that were once considered intractable. Watch for updates on the Cryptiana project, where the remaining unsolved entries may be cleared as model reasoning capabilities improve.
Further reading
Explore ongoing advancements in deep learning research on our Artificial Intelligence section page.
Source note: This article includes information reported by Mail Online.
Live Poll
Do you feel that AI solving long-standing historical puzzles adds value to our understanding of history?
