Cracking a Napoleonic Cipher in Six Hours

Artificial intelligence systems are demonstrating accelerated capabilities in cryptanalysis and mathematical problem-solving, resolving mysteries that have endured for decades and centuries. Recent reports indicate that models from both OpenAI and Anthropic have deciphered complex encrypted material, ranging from 19th-century battlefield communications to foundational digital security algorithms.

In one instance, artificial intelligence engineer Carter Church deployed OpenAI’s GPT-6 Astra model to decode a handwritten military message dating back to 1809. The cipher was used during the Napoleonic Wars by one of Napoleon Bonaparte’s generals and detailed the positions of troops belonging to France and its military allies.

The encrypted letter had remained unexamined in an online archive before Church chose to apply machine learning to the text. While manual cryptanalysis involving the message’s mixture of numbers, letters, and symbols could have demanded weeks of human labor, the AI model completed the task in six hours.

Church stated that the primary advantage demonstrated by the technology was persistence, noting the sense of satisfaction and excitement once the solution materialized. The accuracy of the decryption was independently verified by Satoshi Tomokiyo, a patent engineer who curates an online index of unsolved historical ciphers for cryptographers.

Anthropic Model Analyzes NSA Elliptic Curve Seeds

In a separate cryptographic development, artificial intelligence was applied to modern key generation protocols used to protect digital data across software and hardware. Elliptic curve cryptography relies on specific numerical starting values, commonly known as seeds, to generate cryptographic keys.

For more than 25 years, speculation surrounded five elliptic curve seeds originally created and published by the United States National Security Agency. Observers and critics had long questioned whether the agency designed subtle flaws or backdoor mechanisms into the mathematical values.

An internal Claude model developed by Anthropic analyzed the mathematical origins of all five NSA seeds in less than 30 minutes. The evaluation concluded that the curves contained neither hidden security weaknesses nor concealed jokes. Anthropic noted that while mainstream researchers already placed trust in the standard curves and required no corrective action, the findings represent a decisive closing chapter that demonstrates AI’s analytical reach.

Internal system logs from the model’s reasoning sequence revealed an observation directed at the historical timeline of the problem. During its computation, the model questioned why such an evident mathematical characteristic had eluded human detection for a quarter of a century, underscoring the growing computational efficiency AI brings to complex cryptanalysis.