AI Models Crack Long-Unsolved Enigma Messages
Two cryptanalysts used OpenAI and Anthropic LLMs to decode Enigma messages that had resisted breaking, including one unsolved since 2005.
Two cryptanalysts have used large language models from OpenAI and Anthropic to decode previously unbroken Enigma messages, according to TechCrunch. The breakthroughs target ciphertexts that had resisted conventional cryptanalysis for years.
One developer used OpenAI's Astra model to find and decode an Enigma message that had remained unsolved since 2005. A second cryptanalyst used Anthropic's Claude Opus 5 to break a different unsolved Enigma message.
Retired electrical engineer Frode Weierud validated the solution and maintains the Crypto Cellar website, which holds records of Enigma messages. His verification adds independent confirmation to the AI-assisted decode.
Despite the progress, seven unbroken Enigma messages remain, plus one whose plaintext is known but whose code is unbroken. The Enigma machine was used by Nazi Germany for military communications, and its ciphers were famously attacked by Alan Turing and others at Bletchley Park during World War II.
Quick answers
Which AI models were used to decode Enigma messages?
OpenAI's Astra model and Anthropic's Claude Opus 5 were used by two cryptanalysts to decode separate unsolved Enigma messages.
How many Enigma messages remain unbroken?
Seven unbroken Enigma messages remain, plus one whose plaintext is known but code is unbroken.
Who validated the decoded Enigma message?
Retired electrical engineer Frode Weierud validated the solution and maintains the Crypto Cellar website with Enigma message records.