AI helped decrypt two previously unsolved Enigma messages — TechCrunch

Cryptanalysts said that, using large language models from OpenAI and Anthropic, they decrypted two previously unsolved messages encrypted by Enigma machines. TechCrunch reports.

Astra model's work

Developer Carter Leffen tasked OpenAI's new Astra model with finding an undeciphered Enigma message in a database and decoding it. According to the publication, the model independently conducted archival research, used contextual clues, created an Enigma machine simulator, and reconstructed the plaintext of a message that had puzzled researchers since 2005.

The solution was verified by Frode Weierud, a former electrical engineer interested in cryptology who runs the Crypto Cellar website with a message database and other materials. According to him, Leffen's result filled him with “excitement.”

More current news is available on the UA.News Telegram channel Telegram.

Weierud also noted that Astra's work logs mention archival messages from a “private collection” that are not available on his website. He did not establish whether the model had access to these materials and suggested that another researcher may have published them or that the model may have found them in open archives of the German government.

Second message

On September 21, another cryptanalyst, Jack Willis, informed Weierud that he had used Anthropic's Claude Opus 5 model to decrypt another unsolved message. Willis gave the model more instructions, including using a known signature with the name of a specific officer, which ultimately helped obtain the plaintext.

During World War II, Alan Turing and his team of cryptanalysts created the early Bombe computer, which helped the United Kingdom decrypt Enigma messages. According to Weierud's estimates, seven undeciphered Enigma messages remain today, as well as one for which the plaintext is known but the cipher has still not been broken.