OpenAI GPT-6 Astra deciphered the 1941 Enigma message

The GPT-6 Astra model from OpenAI successfully decrypted an encrypted Enigma message that had remained unsolved since 2005. Researchers used the 1941 historical cipher known as MVUEH to test the analytical capabilities of the neural network in cryptographic analysis.

During the work, the model applied the "crib" method using the word ROSENOW. The key achievement was the detection of a rare left-hand rotor turn, which had not been previously considered by researchers. This step allowed the neural network to restore the machine settings and read the original text of the message, which algorithms had failed to do for almost 20 years.

The results of the experiment confirm the ability of modern LLMs to perform complex logical tasks that require a deep understanding of combinatorics and historical context. The use of GPT-6 Astra allowed for the automation of the state traversal process, which previously required significant time expenditures from expert cryptographers.

This case demonstrates the transition from simple text generation to solving specialized tasks in the field of decryption. The success of the model in working with historical data opens up opportunities for the application of similar architectures in the analysis of encrypted archives, where the human factor and computational limitations have hindered the achievement of results.