Alan Turing, a key figure in computing history, is renowned for developing a test that distinguishes between human and artificial intelligence. Yet, his most significant achievement was decrypting the Enigma code utilized by Nazi Germany during World War II.
Recently, two cryptanalysts announced that they have successfully decoded two previously unsolvable messages encrypted by Enigma machines, thanks to advanced language models created by OpenAI and Anthropic.
During the war, Turing, alongside his team, devised an early computing machine called the Bombe, which enabled the UK to interpret Enigma messages. Nonetheless, a small number of messages from that era remain undeciphered, often due to misrecording or mistakes by the original encoders.
Carter Leffen, a developer, tasked OpenAI’s latest model, Astra, to search through a collection of Enigma messages for an unresolved one and decode it. The model successfully accomplished this after conducting its own research, uncovering contextual clues, simulating the Enigma machine, and eventually revealing the plaintext of a message that had stumped experts since 2005. Leffen also created an interactive website to detail the problem-solving process.
Frode Weierud, a retired electrical engineer passionate about cryptography, operates Crypto Cellar, a website that offers various resources and a database of messages. Last week, Weierud confirmed Leffen’s solution, expressing his amazement at the development.
Interestingly, the logs of the Astra model suggest that it discussed archived messages from a “private collection” not found on Weierud’s site. While Weierud is uncertain if the model accessed these messages, he speculates they may have come from another researcher or possibly public archives from the German government.
“GPT–6 Astra functions like a highly skilled cryptanalyst and archive researcher,” he remarked. “What it achieved in just two days would take a human researcher weeks or even months. I personally spent several weeks examining the Bundesarchiv files that GPT–6 Astra refers to.”
On September 21, another cryptanalyst, Jack Willis, a cybersecurity expert, contacted Weierud to share that he had utilized Anthropic’s Claude Opus 5 model to decipher another unsolved message. Willis provided detailed input, allowing Claude to leverage the known signature of a particular officer’s name to unlock the message.
Weierud points out that only seven Enigma messages remain unresolved, along with one message that has known plaintext, but where the code is still unsolved.
This state of affairs may change soon.
