OpenAI Astra and Claude Opus Crack Two Long-Unsolved Enigma Messages

Two long-unsolved messages encoded with Germany’s Enigma machines have been cracked with help from AI models built by OpenAI and Anthropic, giving modern cryptanalysts new tools for problems that have resisted researchers for years.

Alan Turing and his wartime colleagues famously helped break Enigma traffic using early computing systems, including the Bombe. A small number of archival messages, however, remained unsolved because of problems such as mistranscription and errors in the original encoding.

OpenAI Astra solves a message unsolved since 2005

Developer Carter Leffen asked OpenAI’s Astra model to search a database of unbroken Enigma messages and decode one. According to Leffen’s account, Astra conducted archival research, identified contextual clues, built an Enigma simulator, and eventually recovered the message’s plaintext.

The message had remained unsolved since 2005. Leffen also used Astra to create an interactive site explaining the solution.

Frode Weierud, a retired electrical engineer and longtime cryptology researcher who maintains Crypto Cellar, later validated Leffen’s solution. Weierud wrote that Astra had performed research in two days that could have taken a human researcher weeks or months.

He also noted uncertainty over references in Astra’s logs to archived material in a private collection, saying it was unclear whether the model had accessed those records or found related material elsewhere online or in public German archives.

Claude Opus helps crack a second Enigma message

On September 21, cybersecurity executive and cryptanalyst Jack Willis said he had used Anthropic’s Claude Opus 5 to solve another previously unbroken Enigma message.

Unlike Leffen’s more open-ended Astra experiment, Willis provided Claude with substantially more guidance. The model ultimately used the known signature of a particular officer’s name to help identify the solution, which Weierud documented on Crypto Cellar.

Weierud says seven Enigma messages remain unbroken, along with one additional message whose plaintext is known but whose original code has not yet been reconstructed.