OpenAI GPT–6 Astra breaks Enigma message that has resisted solution since 2005
- Source
- cryptocellar.org
- Author
- sohkamyung
- Date
The report, published on a site dedicated to historical cryptography, describes MVUEH as a wartime Enigma message that had resisted attempts since 2005. According to the researcher, GPT-6 Astra did not call an existing solver. It wrote an Enigma simulator, built Bombe software to search rotor settings, and identified a crib, the guessed plaintext fragment that makes such a search tractable. The result is a single account from one researcher, and the details of how much guidance was given are worth reading in the original before generalizing. Still, it is a concrete example of a model completing a long, multi-stage technical task where each stage depends on the previous one working. That kind of chained tool building is closer to research work than a benchmark score captures.
Context
A validation account by cryptanalyst Frode Weierud reports that OpenAI's GPT-6 Astra recovered the key and plaintext of an Enigma message, a coded German World War II radio transmission, that had resisted solution since 2005. According to the account, the model was only directed to try breaking unsolved messages. It suspected a relationship with a previously solved message, treated a repeated place name as a crib, a guessed stretch of plaintext, and wrote software to search for a compatible machine setting.
The reported achievement combines a textual hypothesis with computational testing. The author says it was immediately clear the correct key and plaintext had been found, while analysis of the model's logs was still ongoing. That supports a specific successful cryptanalytic result without establishing every detail of the claimed autonomous process.
Checking sign-in…
Loading comments…