AI models solve two long-standing code mysteries

Two recent reports describe AI models solving cryptography puzzles that had challenged people for decades or longer. The cases involved a historical military message and the origins of encryption key-generation values.

AI engineer Carter Church used OpenAI’s GPT-6 Astra to decode a message from 1809, written in a code used by one of Napoleon Bonaparte’s generals more than 200 years ago. The letter, found in an online archive, described troop positions belonging to France and its allies. Church said manually decoding its symbols, numbers and letters could have taken weeks; GPT-6 Astra completed the task in six hours. Patent engineer Satoshi Tomokiyo, who maintains an online list of unsolved ciphers, verified the result, according to TechRadar, citing Japan Today.

Separately, an internal Claude model from Anthropic traced the origins of five elliptic curve seeds generated by the US National Security Agency (NSA) in less than 30 minutes, Cyber Security News reported. The seeds had been public for more than 25 years, amid claims they might contain weaknesses usable as backdoors. Anthropic’s model found no weaknesses or inside jokes. Anthropic said researchers largely already trusted the curves and no action was needed, describing the analysis as a “fun closing chapter.”

According to internal thinking logs, Claude also questioned whether it was suspicious that the finding had not been made in 25 years. The reports describe separate research efforts; they do not establish that AI has broadly surpassed human cryptographers.