Astra and Opus just passed Turing’s other test

AI models can now decipher complex historical codes that have long eluded human experts, changing how we can access and interpret historical data.

Two separate teams of cryptanalysts have successfully decoded previously unbreakable Enigma messages from World War II using advanced AI language models. Carter Leffen utilized OpenAI's Astra model, which independently researched historical Enigma messages, built a simulator, and recovered plaintext from a message that had remained undeciphered since 2005. Leffen further developed an interactive website to explain the decoding process. Concurrently, Jack Willis employed Anthropic's Claude Opus 5 model, with more direct guidance, to break another Enigma-encoded message by identifying a specific officer's name signature. These breakthroughs have reduced the number of known unbroken Enigma messages to just seven. The AI models performed these complex tasks in a fraction of the time it would typically take human researchers, demonstrating significant capabilities in historical analysis and code-breaking.

The successful application of large language models like OpenAI's Astra and Anthropic's Claude Opus to crack historically significant and complex codes signifies a new era for AI-driven cryptanalysis. These models are demonstrating an emergent ability to perform deep archival research, contextual reasoning, and simulation building, tasks previously requiring specialized human expertise and extensive time. This development suggests that AI can now be a powerful tool for historical research, potentially unlocking vast amounts of previously inaccessible information encoded in difficult-to-break ciphers.

This advancement impacts fields ranging from historical research and intelligence analysis to cybersecurity. The speed at which these AI models are solving problems that took humans weeks or months suggests a significant acceleration in the pace of discovery and analysis. As the number of unbroken Enigma messages dwindles, the focus will likely shift to what other historical or encrypted data sets these advanced AI capabilities can now tackle, potentially reshaping our understanding of past events and improving future security protocols.