Cryptanalysts have successfully decrypted two long-unsolved Enigma machine messages using advanced language models. Developer Carter Leffen utilized OpenAI’s Astra model to recover the plaintext of a message that had remained unbroken since 2005. The AI conducted archival research, identified context clues, and built a simulator of the Enigma machine to complete the task.
Frode Weirerud, who maintains the Crypto Cellar database, validated the solution. According to Weirerud, the Astra model performed work in two days that would typically require weeks or months for a human researcher. The model’s logs suggest it may have accessed additional private or public German government archives during the process.
Separately, on September 21, cryptanalyst Jack Willis reported using Anthropic’s Claude Opus 5 model to break another previously unsolved message. Willis provided specific guidance, including the known signature of an officer’s name, to assist the model in its analysis.
Currently, there are seven unbroken Enigma messages remaining, plus one message where the plaintext is known but the code itself remains uncracked. These developments highlight the increasing capability of AI agents in performing complex archival research and cryptanalysis tasks.
