AI models crack World War II codes
Summary and headline written by AI from the source article. How we work
Carter Leffen, a developer, used OpenAI’s Astra model to decode a World War II-era Enigma message that had remained unbroken since 2005.
Leffen asked Astra to search for an unsolved message within a database and then decode it. The AI model conducted its own research, simulated the Enigma machine, and successfully recovered the message's original text. Frode Weierud, who maintains a website with Enigma resources, confirmed Leffen’s findings and noted Astra’s speed compared to human researchers.
Astra’s process included examining archived messages from sources Weierud hadn’t hosted, leading to questions about where the AI accessed the information. Weierud believes Astra may have found the messages through another researcher or even German government archives. He described Astra as acting like a “very professional cryptanalyst and archive researcher,” completing work in two days that would take a person weeks or months.
Separately, Jack Willis, a cybersecurity executive, employed Anthropic’s Claude Opus 5 model to break a different unsolved Enigma message on September 21. Willis gave Claude more specific instructions, guiding it to use a known signature to decipher the code. Weierud reports that only seven Enigma messages remain unbroken, plus one where the coded text exists but the original message does not.
