AI models crack two long-standing Enigma cipher messages
Researchers used OpenAI’s Astra and Anthropic’s Claude Opus 5 to decode two Enigma messages that had remained unsolved for years.
Carter Leffen asked OpenAI’s newest model, Astra, to locate and decode an unbroken Enigma transmission, prompting the AI to conduct its own archival search, assemble contextual clues, and simulate the German cipher machine. The effort yielded the plaintext of a message that had eluded scholars since 2005, a result later validated by Frode Weirerud, the maintainer of the Crypto Cellar archive. Weirerud expressed amazement at the speed, noting that the model’s logs referenced a “private collection” of archived messages whose source remains unclear.
On September 21, Jack Willis turned to Anthropic’s Claude Opus 5, providing detailed guidance that enabled the model to exploit a known officer’s name signature and decipher a second long-standing Enigma text. Weirerud reports that just seven encrypted Enigma messages are still unsolved, plus one where the plaintext is known but the cipher remains intact. These breakthroughs suggest AI could soon finish the historic decryption effort that began with Alan Turing’s Bombe.
