What just happened

A Bloomberg developer says he used OpenAI's newest model to decrypt a German military message that had sat unsolved for 83 years. The message, sent through the famous Enigma machine, had resisted every attempt since 1941. Now it's readable — and the person who cracked it is not a cryptographer.

The 83-year-old puzzle

On July 10, 1941, a German soldier used an Enigma machine to encrypt a short radio message — just 82 characters — from the town of Rosenow. He was asking for marching orders and an immediate reply. The intercepted message was filed in an archive of unsolved German communications and sat there, untouched, for decades.

Why brute force was never an option

The Enigma machine offered roughly 159 quintillion possible daily settings. Trying them all one by one was never realistic — even with modern computers. Early attempts failed fast: known keys from the same day didn't fit, and automated searches came back empty.

How the AI actually cracked it

The breakthrough wasn't raw computing power — it was a lucky guess plus a famous Enigma weakness. A nearby message from the same day had already been decrypted, and it contained the town name “Rosenow” twice. The team guessed that name would appear in the unsolved message too, and used it as a search term. Then the weakness: Enigma never encrypts a letter as itself, so any position where the two texts disagreed could be ruled out instantly. At one position, everything lined up — and the remaining 68 characters came out as coherent German: “immediate radio reply” and “specify the march route.”

The honest caveats

First, the AI didn't do it alone. Carter Leffen, a product-development coach at Bloomberg, set the goals and made the calls while GPT-6 Astra and several specialized agents handled the searching, the coding, and the testing — over about ten hours of work. Second, no expert has verified it yet. The published materials don't prove the message's real identity, or that this is the only possible solution. Independent cryptographers still need to check the work. Even the typos in the decrypted text — “BTTE” instead of “BITTE” — are read by Leffen as evidence it's real, on the logic that a fake would more likely be clean.

What it means for you

The story isn't that AI is magic. It's that AI now turns one determined person into a small research team. A non-expert cracked a decades-old cipher in about ten hours by pointing a model at the right archives and letting it search, simulate, and check its own work over and over. That — patience and breadth, not genius — is the real superpower here.

The bottom line

AI didn't out-think the codebreakers of 1941. It just worked through the possibilities faster than anyone had the patience to do alone — and made that patience available to a single person.