Astra Locates and Decodes a 2005 Mystery
Two long-unsolved messages encoded with World War II Enigma machines have reportedly been decoded using large language models from OpenAI and Anthropic. Developer Carter Leffen instructed OpenAI’s Astra to search a database of Enigma messages for an entry that remained unbroken. Leffen claims the model conducted archival research, identified contextual clues, built a simulator of the Enigma machine, and successfully recovered the plaintext of a message that had baffled researchers since 2005. He also utilized Astra to create an interactive website explaining the decoding process.
Validation by Independent Archivist
Frode Weierud, a retired electrical engineer and maintainer of the Crypto Cellar website, validated Leffen’s solution and expressed being “in awe” of the results. Weierud noted that Astra’s logs included discussions of archived messages from a “private collection” not hosted on his site. He speculated that the model may have accessed these files through other online sources, stating, “Personally, I spent several weeks researching the Bundesarchiv files GPT–6 Astra refers to.”
Claude Opus 5 Breaks a Second Message
On September 21, cybersecurity executive Jack Willis reported that he had used Anthropic’s Claude Opus 5 model to decode a different unsolved Enigma message. Willis provided more targeted guidance to Claude Opus 5, which utilized the known signature pattern of a specific officer’s name to recover that message’s plaintext.
Remaining Unbroken Messages
Weierud noted that there are currently seven Enigma messages still unbroken, along with one message where the plaintext is known but the code remains unsolved. The advancements made by Astra and Opus 5 highlight the potential of AI in cryptanalysis.
Original source: TechCrunch AI