OpenAI's GPT-6 Astra Decodes 1941 German Army Enigma Message in Ten Hours

A German Army Enigma message intercepted on July 10, 1941, was decoded this week by Carter Leffen using OpenAI’s GPT-6 Astra model. The solution, now recorded in the CryptoCellar archive, reveals a routine radio request mentioning the place name Rosenow.

How the Break Was Achieved

The model performed roughly 14.8 million key checks over about ten hours of compute time. Leffen supplied a known crib – the repeated place name “Rosenow” from a related message – which reduced the search space to a manageable size. GPT-6 Astra then:

• searched historical archives for contextual clues,
• built an Enigma machine simulator,
• wrote cryptanalysis code,
• tested candidate keys in parallel.

Human Involvement

While the model executed the computational steps, Leffen defined the goal, provided the crib, and oversaw the investigation. Two independent implementations reproduced the same plaintext, confirming the result.

Implications for Modern Cryptography

The Enigma machine, a 1930s rotor cipher, is far less complex than the cryptographic algorithms that secure Bitcoin and other digital assets. Although AI models have identified weaknesses in some encryption systems, the mathematics underlying cryptocurrency wallets remains in a higher security class, and quantum threats to Bitcoin are still an open research question.