OpenAI's GPT-6 Astra Cracks 1941 Enigma Message That Stumped Experts for 20 Years

AI Cracks a 20-Year-Old Enigma Mystery

On September 15, 2026, a message that had defied the world's best cryptanalysts for over two decades finally yielded its secrets. The message, designated MVUEH, was a German Army Enigma transmission from July 10, 1941, that had remained unbroken since it was first published on the Crypto Cellar Research website in 2005. The unlikely solver? OpenAI's GPT-6 Astra, a general-purpose AI model that not only cracked the code but did so with a level of autonomy and sophistication that has left even veteran cryptanalysts in awe.

The breakthrough was first reported by Carter Leffer, who directed GPT-6 Astra to analyze unbroken Enigma messages on the Crypto Cellar Research site. Within two days—September 14 and 15—the AI had achieved what human experts had failed to do for 21 years. Frode Weierud, the cryptanalyst who runs Crypto Cellar Research and maintains the message archive, confirmed the break and published his analysis on September 19, 2026.

The Message That Wouldn't Break

The MVUEH message was sent by a German Army radio station with the tactical callsign 2ny to the SS-Totenkopf Division's Quartiermeister (Ib) station. It was received at 17:30 on July 10, 1941, and logged as message Nr. 172. The 82-letter ciphertext had resisted all decryption attempts since 2005, becoming something of a white whale for Enigma enthusiasts and professional cryptanalysts alike.

Interestingly, a related message from the same day, Nr. 173 (SIPVX), had been broken in 2017 by Alex Shovkoplyas. However, the key recovered for SIPVX differed from the daily Enigma key for July 10, 1941, and crucially, it did not help crack MVUEH. The two messages, it turns out, used completely different keys—MVUEH used wheel order 253 while the daily key and SIPVX used 512. The plug connections and ring settings also differed.

How GPT-6 Astra Did It

What makes this break remarkable is not just the result but the method. According to Leffer's case study, GPT-6 Astra began by analyzing the unbroken messages on the Crypto Cellar site and quickly identified MVUEH as the most promising candidate. It also noticed a potential relationship between MVUEH and the already-broken SIPVX message—a hypothesis that proved correct.

The AI then took a deeply methodical approach. It developed its own Python and C++ implementations of an Enigma simulator and an Enigma Bombe (the code-breaking machine originally developed by Alan Turing and others). After exploring multiple strategies, GPT-6 Astra settled on using the repeated place name ROSENOW ROSENOW as a crib—a known or guessed plaintext segment used to derive the key.

The breakthrough came after running 43,016 batches of the Rosenow search, according to Leffer's case study. The resulting plaintext reads: "Please specify the route of march. I am in Rosenow, Rosenow. Immediate reply by radio," signed by a sender tentatively identified as Waschbusch.

Solving the Riddles Within the Riddles

Weierud's analysis revealed why the message had been so difficult to crack. The original transcription of the ciphertext contained several errors, and—more significantly—the Enigma's left-hand wheel made a turnover at the 72nd letter. This rare event, known as a "left-hand wheel turnover," dramatically complicates decryption because it changes the rotor alignment mid-message.

Additionally, the MVUEH plaintext was nearly identical to the SIPVX message, differing by only twelve letters. This difference was caused by a typo in MVUEH (the word "Bitte" was enciphered as "Btte") and the repetition of the sender's signature "Waschbusch" in SIPVX. These subtle discrepancies masked the relationship between the two messages and likely contributed to the difficulty.

Beyond Code-Breaking: Autonomous Archive Research

Perhaps the most astonishing aspect of GPT-6 Astra's performance was its ability to conduct archival research. In its logs, the AI referenced specific German Bundesarchiv file numbers—RS 3–3/20a and RS 3–3/63b—and mentioned a "private collection" of radio messages. These references were accurate but not available on the Crypto Cellar website, suggesting GPT-6 Astra accessed external resources or databases independently.

Weierud noted in his analysis: "GPT-6 Astra is behaving like a very professional cryptanalyst and archive researcher. What it has achieved in two days would take a human researcher weeks or even months. Personally, I spent several weeks researching the Bundesarchiv files GPT-6 Astra refers to."

A New Era in AI-Assisted Cryptanalysis

This achievement comes amid a flurry of ambitious claims from OpenAI. The same week, the company announced that an internal model had resolved the Navier-Stokes existence and smoothness problem—one of the seven Millennium Prize Problems—though no proof has been published. GPT-6 Astra has also demonstrated impressive performance in other domains, including beating Pokémon FireRed in 18 hours and 12 minutes (versus 96 hours for GPT-5.6 Sol) and completing 60 of 100 harmful robot-arm tasks in safety trials.

However, experts urge caution. The Enigma break, while validated by Weierud, was performed by a user with early access, not by OpenAI itself. The case study is careful to note: "This is not the original breaking of Enigma, and no first-ever or corpus-wide breakthrough is claimed." The Navier-Stokes claim, meanwhile, has no published proof and cannot be independently verified.

Why This Matters

The MVUEH break demonstrates that AI systems can now perform complex, multi-step cryptanalysis with minimal human guidance. GPT-6 Astra didn't just execute a pre-programmed attack—it identified the target, formulated hypotheses, developed its own tools, and adapted its strategy based on intermediate results. This level of autonomous problem-solving has profound implications for cybersecurity, cryptography, and historical research.

For the Enigma community, the break closes a chapter that began in 2005. But for the broader world, it opens a new one: the era of AI as a legitimate research partner in fields requiring deep expertise, creativity, and persistence. As Weierud put it, "The AI break of the Enigma message MVUEH is simply amazing, and as an old cryptanalyst, I am still in awe."

Whether this represents a genuine leap in AI capability or a carefully curated demonstration remains to be seen. But one thing is certain: the message that resisted solution for 20 years has finally been read, and it was an AI that read it.