AI Breakthrough: Astra and Opus Conquer Turing's Other Test!
Modern AI models, including OpenAI's Astra and Anthropic's Claude Opus 5, have successfully cracked two long-unsolved Enigma messages from World War II. This breakthrough highlights the advanced capabilities of Large Language Models in historical cryptanalysis, completing a task that has puzzled researchers for decades.
Computer pioneer Alan Turing is widely recognized for his groundbreaking work, including the famous Turing test designed to assess whether artificial and human intelligence can be distinguished. However, a more pressing challenge he faced during his lifetime was his pivotal role in cracking the Enigma code utilized by Nazi Germany throughout World War II. While Turing and his team of cryptanalysts developed early computing machines, famously dubbed the Bombe, which enabled the United Kingdom to translate Enigma messages during the war, a small number of archival messages have persisted as unbroken, typically due to errors by the original encoders or mistranscription.
In a significant development, a pair of modern cryptanalysts have now reported success in cracking two of these long-unsolved messages, leveraging advanced Large Language Models (LLMs) developed by OpenAI and Anthropic. Carter Leffen, a developer, initiated the process by instructing OpenAI's newest model, Astra, to search a comprehensive database of Enigma messages for an unbroken one and subsequently decode it. The Astra model impressively accomplished this task, performing its own archival research, identifying crucial context clues, constructing a simulator of the Enigma machine, and ultimately recovering the plaintext of a message that had baffled researchers since 2005. Leffen even utilized Astra's capabilities to develop an interactive website that meticulously explains the entire problem-solving process.
Frode Weierud, a retired electrical engineer with a profound and lifelong interest in cryptology, maintains the renowned Crypto Cellar website, which serves as a valuable repository of resources, records, and a database of messages. Last week, Weierud validated Leffen’s solution, expressing his profound