AI models decipher Enigma messages that had remained unsolved for decades

Alan Turing is primarily remembered for the test that bears his name, which seeks to distinguish whether a machine can think like a human being. However, the most decisive challenge he faced in his lifetime was another: deciphering the Enigma code used by Nazi Germany during World War II. Today, two cutting-edge artificial intelligence models seem to be finishing the work that Turing and his team left unfinished.

A pair of cryptanalysts managed to break two messages from Enigma machines that had remained unsolved for years. They did this using OpenAI's Astra models and Anthropic's Claude Opus. While Turing and his collaborators built the Bombe, one of the first computers, to translate those messages in real-time, some archived texts could never be read. In most cases, the problem was a poor transcription or errors from the original encoders themselves.

Carter Leffen, a developer, simply asked Astra to search a database of Enigma messages for any that had not yet been deciphered and to solve it. The model did exactly that. First, it researched files on its own, gathered contextual clues, built a simulator of the Enigma machine, and finally retrieved the plaintext of a message that had puzzled researchers since 2005. It even used Astra to build an interactive website that explains the entire problem.

Frode Weierud, a retired electrical engineer with a lifelong passion for cryptology, maintains the Crypto Cellar site, which gathers resources, records, and a database of messages. Last week, he validated Leffen's solution and said it left him in a state of amazement.

A task that would take a human months

Weierud noted that the model's records include mentions of messages archived in a “private collection” that are not hosted on his site. He is still unsure whether the model actually accessed them or not. He speculates that they may have been shared by another researcher somewhere on the internet, or that Astra could have consulted public records from the German government.

“GPT-6 Astra behaves like a very professional cryptanalyst and archival researcher,” he wrote. “What it achieved in two days would take a human researcher weeks or even months. Personally, I spent several weeks researching the Bundesarchiv files referenced by GPT-6 Astra.”

On September 21, another cryptanalyst, Jack Willis, a cybersecurity executive, contacted Weierud to tell him that he had used Anthropic's Claude Opus model to break a different message. Willis provided the system with much more guidance, which ultimately allowed it to leverage the known signature of a particular officer's name to decipher the text.

Weierud points out that there are only seven Enigma messages left unsolved, in addition to one whose plaintext is already known but whose code has yet to be resolved. Perhaps not for much longer. What Turing started with the Bombe, today's artificial intelligences seem to be finishing with a speed and research capability that leave human specialists breathless.