On October 4, 2026, Carter Church, an engineer at SentinelOne, announced that he used OpenAI’s GPT‑6 model, called Astra, to decipher a Napoleonic‑era code that had been unsolved for over two centuries. The AI completed the job in about six hours after being given just one picture of the cipher and a short text prompt. Church posted the full solution and a script to reproduce the result on his personal blog.
What the AI actually did
Astra is a large language model (LLM) – the kind of AI behind ChatGPT – that can also interpret visual input. Church fed the model a scanned image of the handwritten cipher and asked it to translate the symbols. Within minutes the model produced a working script that turned the image into readable text, and after a few more hours it produced the complete translation of the 24 rows of symbols. The output was enough for historians to confirm the message’s meaning.
Why it matters
For hobby historians and cryptography fans, the result shows that modern AI can tackle very old, hand‑written codes without extensive manual work. It also suggests that similar unsolved historical puzzles might become accessible with a single image and a prompt, though the reliability of AI‑generated translations still needs expert verification.
