For 217 Years, Cryptographers Couldn’t Decode Napoleon’s Secret Letter. Then an AI Cracked It in 6 Hours.

3 min readHere’s what you’ll learn when you read this story:A cybersecurity professional used an OpenAI model to solve a cipher embedded in a letter sent on Napoleon Bonaparte’s orders in 1809.The 217-year-old letter was written on the eve of war and laid out the positions of French and allied armies, all in code.OpenAI’s GPT-6 Astra cracked the code using an unusual multidisciplinary approach.Forget the Rosetta Stone, the Enigma machine, or even Bletchley Park’s Colossus: AI may now offer all that’s needed to crack historical codes. Carter Church, an AI engineer at cybersecurity firm SentinelOne, provided the example, using OpenAI’s GPT-6 Astra to decipher a coded letter sent to one of Napoleon’s generals in 1809. All he supplied was a single prompt and a scanned image; about six hours later, he had the answer.“What makes this impressive isn’t actually the codebreaking, but that Astra completed the entire multi-modal workflow in [about] six hours from a single image and goal,” Church wrote in a post on X. In that time, he added, Astra transcribed 1,300 hand-drawn signs, found a partial key that had never been connected to the letter, cracked the rest with its own simulated annealing solver, and cross-referenced surrounding letters to correct the document’s published date.As Church explained in a full write-up of the exercise, the letter opens with a single line of plain French and then jumps straight into 24 rows of numbers, letters, and invented symbols. “It sat on Cryptiana’s Unsolved Historical Ciphers list for decades,” he wrote—and the site’s editor has since marked the entry as solved. “It’s a war briefing sent two weeks before Austria invaded.”The chain of events behind that briefing began in Paris. On March 16, 1809, Napoleon wrote to his stepson Eugène, the Viceroy of Italy, instructing him to send a letter to General Auguste de Marmont, who would not receive his marshal’s baton until later that year. Marmont’s troops sat in Dalmatia, isolated from every other French army—a dire circumstance should war with Austria break out. Napoleon wanted the letter to spell out the positioning of every French and allied army, and he told Eugène to send it by an intelligence officer and to put it in cipher in case it fell into the wrong hands. The enciphered letter that Astra read is the surviving result of that order.Breaking it was not a single task but several. Church wrote that the letter used a homophonic cipher, in which common letters are represented by several interchangeable signs. Working from the only publicly available image—a low-resolution reproduction published in a 1969 French Army journal—the model divided the plate into rows and worked out which of the 1,300 hand-drawn marks were repeat instances of the same sign, 155 distinct signs in all, before breaking the cipher itself. “Transcription and cryptanalysis are usually separate skills, often separate careers,” Church wrote. “Here they ran as one workflow, and the plate stayed in the loop: proposed values were checked back against every occurrence of a sign on the image.”The end product, he wrote, was “the same object the clerk in Eugène’s office had on his desk in March 1809.”Even so, the result took some getting used to. “I didn’t believe it at first,” Church told Live Science. “Everyone has the experience of seeing a model claim something outrageous, but the difference here is that the solution is mechanically testable: there’s essentially zero chance an incorrect key could produce coherent and historically consistent output.”Church was careful to note that the cipher itself wasn’t the tricky part, and that cryptographers could have cracked it “easily” had they devoted the time and energy to do so. What made the letter sit unread, he argued, was the need to transcribe, catalog, translate, and then crack it—a multidisciplinary stretch few specialists can afford. “Every field has an unsolved backlog of niche problems that specialists don’t have the bandwidth to tackle,” he wrote. “The takeaway here is that the people shortening those lists will increasingly be outsiders who were simply curious.”Tim Newcomb is a journalist based in the Pacific Northwest. He covers stadiums, sneakers, gear, infrastructure, and more for a variety of publications, including Popular Mechanics. His favorite interviews have included sit-downs with Roger Federer in Switzerland, Kobe Bryant in Los Angeles, and Tinker Hatfield in Portland.