AI Helps Crack Two Long-Unsolved Enigma Messages After Decades of Mystery
Two Enigma-coded messages that had resisted researchers for years have reportedly been deciphered with assistance from advanced AI models, marking a fascinating moment for historical cryptanalysis and digital archival research.
The breakthroughs involved OpenAI’s GPT-6 Astra and Anthropic’s Claude Opus 5, both used in separate attempts to solve messages encrypted by the legendary Enigma machine. The results were reviewed by veteran Enigma researcher Frode Weirerud, who confirmed the solutions and highlighted how artificial intelligence can dramatically accelerate research that once took weeks, months, or even longer.
The Enigma machine remains one of the most famous encryption systems in history. Used by Nazi Germany during World War II, it became closely linked with Alan Turing and the British codebreaking teams at Bletchley Park. Turing and his colleagues developed the Bombe, an electromechanical device that helped decrypt huge volumes of German military communications and played a major role in the Allied war effort.
Even so, not every Enigma message was solved. Some archived messages remained unreadable for decades, often because of missing information, transcription errors, incorrect settings, or mistakes made by the original operators. These unsolved fragments have continued to intrigue historians, cryptographers, and puzzle enthusiasts.
The first recent success came from developer Carter Leffen, who reportedly asked GPT-6 Astra to locate an unresolved Enigma message from historical archives and attempt to decrypt it. Rather than simply guessing, the AI model worked through the problem like a research assistant. It examined the historical context, searched for clues, created its own Enigma simulator, tested possibilities, and eventually recovered the plaintext of a message that had remained unsolved since 2005.
Leffen also used Astra to build an interactive presentation explaining the mystery, the decoding process, and the proposed solution. This helped make the technical work easier to follow for others interested in cryptography and World War II history.
Frode Weirerud, a retired electrical engineer and longtime Enigma specialist, later reviewed the work and confirmed the solution. He praised the AI’s performance, saying it acted much like a skilled cryptanalyst and archive researcher. According to Weirerud, the work completed in just two days might have taken a human researcher weeks or even months.
However, he also noted one important uncertainty. Some of the information referenced by the AI appeared to involve material from a private collection, and it was unclear whether the model had found those details through public sources, other online material, or German archival records. That question shows why human verification remains essential when AI is used for historical research.
A second Enigma message was decrypted only days later. On September 21, cryptanalyst Jack Willis informed Weirerud that he had used Claude Opus 5 to solve another message that had remained unread for years. In this case, the AI received more direct help from a human expert. Willis supplied key guidance, including the known signature of a specific officer, which gave the model a valuable clue and helped lead it toward the correct solution.
These achievements do not mean artificial intelligence can now break Enigma messages entirely on its own. The first case involved a combination of AI-driven research, programming, simulation, and reasoning. The second relied heavily on expert direction and historical knowledge. Still, both examples show that AI can become a powerful tool for cryptanalysis, especially when paired with experienced researchers.
For historians and codebreaking enthusiasts, the discoveries are exciting because they suggest that other long-forgotten encrypted messages may eventually be solved. AI can rapidly search archives, build test tools, identify patterns, and explore possibilities at a pace that would be difficult for humans working manually.
Weirerud says seven Enigma messages remain unsolved, along with one message where the plaintext is known but the exact encrypted solution has not yet been reconstructed. With modern AI now entering the field of historical codebreaking, those remaining mysteries may not stay hidden forever.
