Astra and Opus crack Enigma messages unsolved for decades

Alan Turing is remembered for the test that carries his name, but his hardest real-world problem was breaking Germany's Enigma cipher during the war. Two AI models have now solved intercepted Enigma messages that had gone unbroken for years. Developer Carter Leffen pointed OpenAI's Astra at a message that had defeated researchers since 2005. The model did its own archival research to work out the likely context, wrote a working Enigma simulator, and recovered the plaintext. On September 21, cybersecurity executive Jack Willis used Anthropic's Claude Opus 5 to break a separate message, giving it one clue: an officer's name that was likely to appear in the signature.

Frode Weierud, the cryptanalyst who maintains the main catalog of unsolved intercepts, checked both results and confirmed them. He said what Astra managed in two days would take a human researcher weeks or months. Only seven unsolved messages from that collection now remain, so the machines may finish the job before long. TechCrunch has the full account of both breaks.

Why it matters

This is a clean, checkable example of an agent doing real multi-step research: finding sources, forming a hypothesis, writing and running its own code, and landing on an answer that an expert can verify against ground truth. If you are trying to judge what today's agents can actually do, that loop matters more than a benchmark score.