Key facts
- AI models Astra and Claude Opus have cracked two long-unsolved Enigma-encoded messages.
- Astra performed archival research, built an Enigma machine simulator, and recovered plaintext.
- Claude Opus used the known signature of an officer's name to break a message.
- These achievements would typically take human researchers weeks or months.
- Only seven unbroken Enigma messages remain, plus one with known plaintext but unbroken code.
Two advanced artificial intelligence language models have successfully decoded two long-unsolved messages encoded by Enigma machines, achievements that cryptanalysts say would typically take human researchers weeks or months.
Developer Carter Leffen utilized OpenAI's Astra model to search a database of Enigma messages, ultimately finding and decoding a message that had baffled researchers since 2005. The AI model reportedly conducted its own archival research, identified contextual clues, built a simulator of the Enigma machine, and recovered the plaintext. Leffen also used Astra to create a website explaining the problem.
Frode Weirerud, a retired electrical engineer and cryptology enthusiast who maintains the Crypto Cellar website, validated Leffen's solution, describing it as awe-inspiring. Weirerud noted that Astra's logs indicated the model may have accessed archived messages from a "private collection" not hosted on his site, speculating they might have been shared online or accessed from German government archives. He commented that Astra behaved like a "very professional cryptanalyst and archive researcher."
Separately, cryptanalyst Jack Willis reported using Anthropic's Claude Opus 5 model to break a different unsolved Enigma message. Willis provided more specific guidance to Claude, enabling it to use the known signature of a particular officer's name to decipher the message.
According to Weirerud, there are now only seven Enigma messages remaining that are still unbroken, along with one message where the plaintext is known but the code itself has not been deciphered.