Researchers are increasingly using frontier LLMs like GPT-6 Sol and Opus 5.5 to solve historical problems, particularly in cryptography, textual tracing, and cross-disciplinary knowledge synthesis. Early results show promise in decoding 17th-century alchemical texts and identifying previously unknown translations, suggesting that collaborations between historians and AI labs could produce meaningful advances in historical knowledge.
GPT-6 Astra autonomously decrypted an unbroken Enigma message (Nr. 172, MVUEH) by analyzing historical ciphertexts, identifying a repeated place name crib (ROSENOW), and developing custom Enigma simulator and Bombe software. Carter Leffer prompted the AI to attempt the task, but the system independently selected the target and executed the cryptanalysis.
On September 15, 2026, Carter Leffer validated OpenAI's GPT-6 Astra breaking a German Army Enigma message (MVUEH) from July 10, 1941, that had resisted solution since 2005. GPT-6 Astra autonomously analyzed unbroken messages, identified MVUEH as promising, and developed Enigma simulator software to execute a successful break using the crib ROSENOW, discovering the correct key and plaintext.
Researchers successfully decrypted a German Army Enigma message from July 10, 1941, requesting a route of march and immediate radio response. The message was recovered by identifying repeated plaintext phrases and machine settings, revealing the sender's location as Rosenow and signature as Waschbusch.