Developer Uses Astra to Decode an Enigma Message Unsolved Since 2005

Carter Leffen gave OpenAI’s Astra a broad assignment: find an unsolved Enigma message and decode it. The model recovered the plaintext of a message that had resisted researchers since 2005, and cryptologist Frode Weierud validated the solution. But one part of Astra’s research trail remains unclear: whether it accessed messages from a private collection.

From an open-ended request to plaintext

Leffen, a developer, asked Astra to search a database of Enigma messages for one that remained unbroken. He did not supply a specific message to decode. According to TechCrunch, Astra researched archival material, identified context clues and built a simulator of an Enigma machine before arriving at the plaintext.

That sequence matters because this was not just a request to read an encrypted message. Astra had to find a candidate, gather information that might help interpret it and test a possible decoding with a machine simulator. Leffen also used Astra to build an interactive website explaining the problem.

The human check

Weierud maintains Crypto Cellar, a collection of cryptology resources and message records. He validated Leffen’s solution the week before TechCrunch’s September 25 account and said it left him in “awe.” His check gives the decoded result a stronger footing than a model’s answer alone; it does not settle every question about the route Astra took to reach it.

Enigma messages were encoded by Nazi Germany during World War II. A small number of archival messages have remained unbroken, often because a transcription is wrong or an original encoder made an error. Those possibilities make historical records and surrounding clues important, not merely the ability to simulate the machine.

What it has achieved in two days would take a human researcher weeks or even months.

Frode Weierud, writing about Astra’s work

An unanswered question in the logs

Astra’s logs discuss archived messages in a “private collection” that Weierud does not host. He cannot tell whether the model accessed those messages. He suggested two possibilities for where related material might have come from: another researcher may have shared it online, or Astra may have reached public German government archives.

The distinction is between checking an answer and accounting for all the material used to produce it. Weierud confirmed this solution, while the collection mentioned in the logs remains an open question. His count of the remaining work is precise: seven Enigma messages are still unbroken, plus one whose plaintext is known although its code has not been broken.