GPT-6 Astra cracks a 217-year-old Napoleonic cipher in six hours

Started by BigDogCena41, Today at 08:23 PM

Previous topic - Next topic

0 Members and 1 Guest are viewing this topic.

Topic: GPT-6 Astra cracks a 217-year-old Napoleonic cipher in six hours   Views(Read 80 times)
Active members in this topic:
BigDogCena41(1)

BigDogCena41

OpenAI's GPT-6 Astra has decrypted a 217 year old Napoleonic cipher in about six hours, working from a single image and one prompt, Tom's Hardware reports. The attempt was run by Carter Church, an engineer at cybersecurity firm SentinelOne. The cipher was an unsolved entry in Cryptiana, a database of historical ciphers. Its ciphertext ran to 24 rows of "numbers, letters and invented symbols"

The decrypted text turned out to contain military orders from Eugène de Beauharnais, Napoleon's stepson, to French General Auguste de Marmont about preparations for Napoleon's 1809 campaign against Austria. It even fills a gap in Napoleon's published correspondence from 1865, where a sentence had previously trailed off into an ellipsis. Church wrote: "I used GPT-6 Astra to break an unsolved cipher to one of Napoleon's generals that had gone unread for 217 years."

The method is interesting. According to Church, Astra used its ability to read images to find a partial key, and then "cracked the rest with its own simulated-annealing solver." In other words, the model wrote and ran its own code to finish the job. The result is "mechanically testable against the ciphertext", which means anyone can check that the decryption works. Satoshi Tomokiyo, who maintains the database, has verified it and marked the cipher as solved

This is the kind of AI story that is easy to enjoy. Nobody is harmed, history gains a little, and the result can be checked independently. It also shows how much AI systems have changed. A few years ago, chatbots struggled with simple substitution ciphers. Now one can read a handwritten table of invented symbols, form a hypothesis and write a solver in an afternoon

Would you trust an AI decryption of a historical document without a human expert checking it? And which famous unsolved cipher should AI tackle next?