AI Models Crack Two Unsolved WWII Enigma Messages, Leaving Only Seven
News Summary
Two of the most capable AI systems available today, OpenAI's newest model Astra and Anthropic's Claude Opus 5, have each helped researchers decipher a World War II Enigma message that had resisted human cryptanalysts for years. The results, reported by TechCrunch on September 25, 2026 (Eastern Time) and echoed by several aggregators, were checked by an independent expert in the field, and they offer a vivid example of how modern AI can support historical research and cryptology education.
Why This Is Called Turing's "Other Test"
Alan Turing is best known for the imitation game that now bears his name, a thought experiment about whether a machine can converse like a person. During his lifetime, however, a more consequential challenge was breaking the Enigma cipher used by Germany in World War II. Enigma machines scrambled text with rotating wheels and plugboard wiring, and the number of possible settings was enormous. The headline framing treats the decryption of surviving, unsolved Enigma messages as a second, more practical benchmark for machine reasoning.
Astra Cracks a Message Unsolved Since 2005
According to the reporting, developer Carter Leffen asked Astra to search a database of Enigma messages, pick one that had never been broken, and decode it. Rather than simply running brute-force guesses, the model reportedly carried out its own archival research and looked for contextual clues about where and when the message was likely produced. It then built a simulator of the Enigma machine and used it to recover the plaintext. The message had baffled researchers since 2005. Leffen's work took place in September 2026.
Claude Opus 5 Breaks a Different Message
On September 21, 2026, cybersecurity executive and cryptanalyst Jack Willis contacted cryptology researcher Frode Weierud to say he had used Anthropic's Claude Opus 5 to break another unsolved message. Willis gave the model considerably more guidance than Leffen gave Astra. Claude then used the known signature of a particular officer's name, a typical crib in Enigma work, to recover the message.
Independent Validation
Weierud, a retired electrical engineer who maintains the Crypto Cellar, a well-known community resource for Enigma research, validated both solutions. He described the OpenAI model as behaving like a very professional cryptanalyst, and noted it completed work that would take a human researcher weeks or even months. Reports also mention that only seven unbroken Enigma messages remain in the collection, which makes these two solutions a meaningful share of what was left.
What Differed Between the Two Approaches
The two cases show different levels of autonomy. Astra was given a high-level goal and handled the research, tooling and decryption steps itself. Opus 5 worked in a more collaborative loop, with a human supplying hints and direction. Both patterns are common in practical AI-assisted research: some tasks can be delegated end to end, while others benefit from an expert steering the model.
Why It Matters for Science and Education
The episode is a compact case study in how AI can combine information retrieval, hypothesis testing and software construction. Building a working machine simulator on demand, then using it to test candidate settings, is the kind of multi-step problem solving that students of mathematics, computer science and history can learn from. It also highlights the value of human verification: the claims carry weight because a domain expert independently confirmed them.
What to Watch Next
With only a handful of unbroken messages left, attention is likely to turn to whether similar methods can help with other historical ciphers and archives. Readers should treat details beyond the confirmed reports, including exact model configurations and the number of attempts, as not yet publicly documented.