Two cryptanalysts have reported successfully decoding long-unsolved Enigma messages using large language models (LLMs) from OpenAI and Anthropic.
Model ReleasesOpenAIAnthropicAstraClaude Opus 5
OpenAI's Astra and Anthropic's Claude Opus 5 Decode Long-Unsolved Enigma Messages
Developer Carter Leffen used OpenAI’s Astra model to recover the plaintext of a message that had remained unbroken since 2005. The model reportedly conducted archival research, utilized context clues, and built a simulator of the Enigma machine to achieve the result. Frode Weierud, a retired electrical engineer and maintainer of the Crypto Cellar website, validated the solution. Weierud noted that the model's performance was remarkably efficient, performing in two days what would take a human researcher weeks or months.
Separately, cybersecurity executive Jack Willis reported that Anthropic’s Claude Opus 5 model successfully broke a different unsolved message. According to Willis, the model utilized the known signature of a specific officer's name to achieve the decryption with significant guidance.
Cryptologists note that only seven unbroken Enigma messages remain in existence. Weierud commented that the model's capabilities suggest it is behaving like a "professional cryptanalyst and archive researcher."
Sources
- Astra and Opus just passed Turing’s other test （TechCrunch AI, 2026-09-25）