Astra and Opus, developed by OpenAI and Anthropic respectively, cracked two long-unsolved Enigma messages, one of which had baffled researchers since 2005. The models used advanced decryption techniques to decode messages encoded by Enigma machines during World War II. The breakthrough was achieved by leveraging large language models to search through databases and build simulators of the Enigma machine.
Astra, OpenAI’s newest model, was able to decode an unbroken message after conducting its own archival research, finding context clues, and building a simulator of the Enigma machine. The model also helped create an interactive website explaining the problem, which was validated by Frode Weierud, a retired electrical engineer and cryptology enthusiast.
Weierud noted that Astra’s logs included discussion of archived messages in a 'private collection' not hosted by him, raising questions about how the model accessed these resources. He speculated they may have been shared by another researcher online or accessed from the German government’s public archives. He described Astra as behaving like a 'very professional cryptanalyst and archive researcher'.
On September 21, another cryptanalyst, Jack Willis, used Anthropic’s Claude Opus 5 to break a different unsolved message. Willis provided significant guidance to the model, which used the known signature of an officer’s name to decode the message. Weierud highlighted that only seven unbroken Enigma messages remain, with one where the plaintext is known but the code is still unbroken.
The models’ success has sparked interest in the remaining Enigma messages, with Weierud noting that the breakthrough may not last long. The achievements of Astra and Opus demonstrate the potential of AI in historical cryptanalysis and archival research. The models have shown capabilities that would take human researchers weeks or even months to achieve.
Source: techcrunch
