AI Takes Napoleon's Code Apart in Just 6 Hours

Napoleon's encrypted war orders just got cracked by an AI model that needed little more than a blurry scan and six hours. Carter Church, an engineer at cybersecurity firm SentinelOne, used OpenAI's experimental GPT-6 "Astra" to decode a 1809 letter written in cipher by Napoleon's stepson Eugene to Marshal Auguste de Marmont, detailing French and allied troop positions across Europe, Live Science reports. Napoleon instructed Eugene to send the encrypted message to the commander, who was preparing for an Austrian invasion in what is now Croatia, reports the Times of London.

The letter "opens with one line of plain French, then 24 rows of numbers, letters and invented symbols," Church said on X. The model first transcribed 1,300 symbols from the low-resolution image, then independently surfaced a decades-old partial code table compiled by historian Daniel Tant, which had not been previously linked to the letter, and used it to finish the key.

• The content largely matches known correspondence and offers no bombshell revelations, though historian Zack White, host of the Napoleonic Wars Podcast, says it raises "some curious questions about Napoleon's mindset in the run up to the campaign in Austria in 1809." The letter overstates the number of troops available by 60,000 and incorrectly claims that "the Russians are marching on Austria." It also includes some words of encouragement: "When you receive the order to march, you must not allow yourself to be intimidated by a few troops or a gathering of rabble."

Church says human cryptographers could have cracked the cipher with enough time and effort, but the impressive part is how Astra completed the "entire multi-modal workflow" in six hours "from a single image and goal." Michael Rowe at King's College London says that because Napoleon detailed what the encrypted message should include in a letter to his stepson, researchers can be reasonably confident that the translation is accurate, and that the method will work with other ciphers.

• "It's a bit like a kind of Rosetta Stone," Rowe tells Live Science. "If you've tested out AI on this letter, where you pretty much know what the contents would have been because it's referred to in another letter, the AI does its thing and essentially reproduces what you're expecting to see. Then if you use the exact same kind of tool and methodology on something where you don't know [the contents], you're more likely to trust the outcome."