AI Just Cracked WWII Enigma Messages Humans Couldn't Solve for 20 Years

During World War II, Alan Turing and his team built an early computer to crack Enigma, the coded messages Nazi Germany used to send military orders. They succeeded, but a handful of messages were never decoded — usually because of typos or mistakes by the original German operators. Those leftovers have sat unsolved in archives for decades, and until recently, only a few specialists in the world could even attempt them.

Then a developer named Carter Leffen simply asked OpenAI's newest model, Astra, to find and decode one. Without step-by-step instructions, the AI researched the archives, picked up clues, built a working simulation of the Enigma machine, and recovered the message — one that had stumped researchers since 2005. It even built a website explaining the solution. A retired engineer named Frode Weierud, who has kept an Enigma archive for years, checked the answer and confirmed it was real. Weeks later, a cybersecurity executive used Anthropic's Claude Opus to break a second message using a different approach.

Why should you care? Two things. First, this is a clear demonstration of AI doing original research, not just repeating things it read — the kind of multi-step problem-solving that used to require a trained human expert. Second, it hints at what these tools mean for secrets and security. Encryption is what protects your bank logins, medical records, and private messages. The AI here was attacking an outdated 1940s cipher, not modern encryption, so your passwords are not suddenly at risk. But the pattern — AI grinding through hard problems faster and cheaper than people — is the part worth noticing.

There is a catch, and it is a strange one. Weierud cannot fully explain how Astra found its answer. The AI's logs mention archived messages in a "private collection" that was not in the database it was given. He suspects the model found them posted somewhere online or pulled them from a public German government archive, but he cannot say for certain. That is a real concern: when AI solves a puzzle, it does not always show its work in a way humans can verify. Weierud estimates the task would take a human researcher weeks or months. Only seven unsolved Enigma messages remain.