Astra and Claude Opus 5 crack two long-unsolved Enigma messages

Alan Turing is best known for the test that carries his name, but the harder problem he faced in his lifetime was breaking the Enigma code used by Nazi Germany. Now two cryptanalysts say they have cracked long-unsolved Enigma messages with the help of LLMs built by OpenAI and Anthropic, TechCrunch reported on September 25.

Turing's team built an early computer called the Bombe to translate Enigma traffic, yet a handful of archived messages stayed unbroken, usually because of mistranscriptions or errors by the original encoders.

Developer Carter Leffen simply asked OpenAI's newest model, Astra, to search a database of Enigma messages for an unbroken one and decode it. The model did its own archival research, found context clues, built a simulator of the Enigma machine and ultimately recovered the plaintext of a message that had baffled researchers since 2005. Leffen also used Astra to build an interactive website explaining the problem.

Frode Weirerud, a retired electrical engineer who maintains the cryptology site Crypto Cellar and its database of messages, validated Leffen's solution last week and said it left him in "awe."

One detail stands out for anyone tracking how far agents will go: Astra's logs discuss archived messages in a "private collection" that Weirerud does not host. He still is not sure whether the model accessed them, and speculates they may have been shared by another researcher online or that the model reached the German government's public archives.

Weirerud's assessment of the speed was blunt. He wrote that what the model achieved in two days would take a human researcher weeks or even months, and that he personally spent several weeks researching the Bundesarchiv files the model refers to.

On September 21, another cryptanalyst, Jack Willis, reached out to Weirerud to say he had used Anthropic's Claude Opus 5 to break a different unsolved message. Willis gave the model significantly more guidance, and it was ultimately able to use the known signature of a particular officer's name to complete the break.

Weirerud notes that just seven unbroken Enigma messages remain, along with one message whose plaintext is known but whose code is still unbroken. For a small community that has chipped away at these messages for decades, the backlog is now clearing at model speed.