Astra and Opus, advanced large language models developed by OpenAI and Anthropic respectively, have recently achieved a remarkable feat by decoding previously unsolved World War II messages encrypted with the German Enigma machine. These messages had eluded human cryptanalysts for decades, often due to errors or mistranscriptions in the original encoded texts. This breakthrough not only demonstrates the models’ impressive analytical capabilities but also honors the legacy of Alan Turing, who famously worked to break the Enigma code during the war.

Carter Leffen, a software developer, used OpenAI’s Astra to identify and decode one such long-unresolved Enigma message. Astra autonomously conducted archival research, built an Enigma machine simulator, and extracted the plaintext from a message that had baffled cryptologists since 2005. Leffen even created an interactive website to explain the decoding process. His solution gained validation from Frode Weierud, a retired engineer and cryptology enthusiast who maintains a comprehensive Enigma message database on his website Crypto Cellar.

Shortly after, Jack Willis, a cybersecurity executive and cryptanalyst, employed Anthropic’s Claude Opus 5 model to crack another unsolved Enigma message. With more direct guidance, Claude leveraged known officer signatures in the message to arrive at the plaintext, marking another significant success for AI-driven codebreaking. Currently, only a handful of Enigma messages remain indecipherable, and these breakthroughs suggest that artificial intelligence may soon unlock those final mysteries.

The use of Astra and Opus to decode these historic messages highlights the growing role of AI in complex problem solving, surpassing what even dedicated experts can achieve in short timespans. Both models demonstrated sophisticated investigative skills, akin to professional cryptanalysts, by accessing diverse sources and piecing together contextual clues. This convergence of cutting-edge AI technology and historical research could close the chapter on lingering Enigma enigmas and inspire broader applications for AI in cryptography and archival studies.