AI Cracks Two Unsolved WWII Enigma Messages That Stumped Researchers for Decades

AI Achieves What Human Cryptanalysts Couldn't for Decades

Two advanced AI models have independently cracked unsolved Enigma messages from World War II, demonstrating capabilities that extend far beyond simple language processing into autonomous research and cryptanalysis1

1

. OpenAI's GPT-6 Astra and Anthropic's Claude Opus 5 each decoded different historical ciphers that had baffled researchers for years, marking a significant milestone in AI's ability to tackle complex, real-world problems without extensive human guidance.

Developer Carter Leffen simply asked Astra to search a database of Enigma messages for an unbroken message and decode it1

1

. The autonomous AI selected the MVUEH message, an 82-letter German Army transmission sent to the SS-Totenkopf Division on July 10, 1941, which had remained unsolved since being shared online in 20052

2

. What Astra accomplished in two days would have taken human researchers weeks or even months, according to cryptology expert Frode Weirerud, who maintains the Crypto Cellar website containing Enigma message databases1

1

.

Building Tools to Crack the Code

The most remarkable aspect of Astra's achievement lies in its autonomous approach to cryptanalysis. The AI didn't just attempt to decode the message through brute force or pattern matching. Instead, it conducted archival research, identified context clues, and developed its own Python and C++ software for an Enigma simulator and Enigma Bombe—the same type of early computer that Alan Turing and his team built during World War II to break German codes2

2

.

Astra identified the MVUEH message as a promising target and hypothesized it might be related to another message, SIPVX. Expecting the repeated place name "ROSENOW ROSENOW" as a probable plaintext clue, the AI successfully decrypted the 82-letter cipher2

2

. The decrypted German Army transmission read: "Please specify the route of march. I am in Rosenow, Rosenow. Immediate reply by radio."

Professional-Grade Research Capabilities

Weirerud, who validated Leffen's solution, expressed awe at the AI's performance. "GPT-6 Astra is behaving like a very professional cryptanalyst and archive researcher," he wrote1

1

. The model's logs revealed it discussed archived messages in a "private collection" not hosted by Weirerud, though it remains unclear whether the AI actually accessed them or found them shared by another researcher online. Weirerud speculated the model might have accessed the German government's public archives, noting he personally spent several weeks researching the same Bundesarchiv files that Astra referenced1

1

.

This raises important questions about AI agents' ability to navigate and access information sources autonomously. The lengths to which these models will go to answer questions demonstrates both their potential value and the need to understand their research methodologies.

Claude Opus 5 Breaks Second Message

On September 21, cryptanalyst Jack Willis contacted Weirerud with news that he had used Claude Opus 5 to break a different unsolved Enigma message1

1

. While Willis provided significantly more guidance to Claude compared to Astra's autonomous approach, the AI successfully used the known signature of a particular officer's name to break the message1

1

. This demonstrates that both autonomous and guided AI approaches can achieve breakthroughs in cryptanalysis.

Understanding Enigma's Complexity

The Enigma machine's sophistication made it one of history's most challenging ciphers to crack. The device featured a 26-key keyboard with a lampboard above it, three rotors selected from a set of five, and a reflector that returned signals2

2

. The rotors advanced with every keypress, meaning repeatedly pressing the same letter would illuminate different letters each time. A plugboard at the front added another layer of obfuscation, allowing operators to use up to 10 wires to swap letters around2

2

.

Operators needed to configure four settings before use: rotor order, ring setting, rotor starting positions, and plugboard connections. These settings changed daily according to a calendar distributed separately, meaning even possessing an Enigma machine wasn't enough without knowing the day's specific configuration2

2

.

What This Means for AI and Historical Research

The successful decryption of these unsolved Enigma messages represents more than just a historical curiosity. It demonstrates AI's capability to tackle complex, multi-step problems that require research, tool creation, hypothesis formation, and iterative testing. Astra recently cracked a 108-year-old unsolved WWI German code encrypted using the ADFGVX method, showing its cryptanalysis skills extend beyond World War II ciphers2

2

.

Weirerud notes that only seven unbroken Enigma messages now remain, along with one message where the plaintext is known but the code remains unsolved1

1

. Given Astra and Opus's demonstrated capabilities, these remaining historical ciphers may not stay mysterious much longer. The AI models have effectively passed Turing's other test—not distinguishing human from machine intelligence, but cracking the Enigma codes that Turing himself worked to defeat during World War II1

1

.

Watch for AI models to increasingly contribute to historical research, archaeology, and other fields where pattern recognition, archival research, and complex problem-solving intersect. The autonomous nature of these breakthroughs suggests we're entering an era where AI can serve as genuine research partners rather than just tools that execute predefined tasks.