AI models crack decades-old Enigma codes that stumped human cryptanalysts

Large language models from OpenAI and Anthropic have achieved what human cryptanalysts could not: deciphering Enigma messages that have remained unbroken for decades. The breakthroughs highlight AI's capacity to combine archival research, pattern recognition, and technical problem-solving in ways that accelerate complex investigation work.

Two Machines, Two Victories#

Developer Carter Leffen tasked OpenAI's Astra model with locating and decoding an unsolved Enigma message from the historical record. The system conducted its own research across cryptographic databases, identified contextual clues, simulated an Enigma machine's mechanics, and ultimately recovered plaintext from a message that had resisted decryption since 2005. Leffen then used the same model to create an interactive educational website documenting the entire process.

Days later, cryptanalyst Jack Willis employed Anthropic's Claude Opus 5 to crack a second unsolved message by providing targeted guidance—the model identified a known military officer's signature and used it as a cryptographic key to break the code.

Verification and Astonishment#

Both solutions have been validated by Frode Weirerud, a retired electrical engineer who maintains Crypto Cellar, an archive of Enigma materials and unsolved messages. Weirerud expressed profound respect for the achievement, noting that Astra's work "would take a human researcher weeks or even months." He personally spent several weeks studying the same German government archives that Astra consulted.

What Remains#

Weirerud estimates seven unsolved Enigma messages still exist in the historical record, plus one message whose decrypted content is known but whose encoding mechanism remains undetermined. The speed and effectiveness of the AI approach suggests those remaining puzzles may not stay unsolved much longer.

The episode underscores how modern language models function less as mere chatbots and more as capable research assistants, autonomously pursuing complex, multi-step investigative tasks across dispersed archival sources.