OpenAI GPT-6 Astra and Anthropic Claude AI Break Historical Ciphers and Verify NSA Cryptographic Seeds

OpenAI’s GPT-6 Astra model has successfully decrypted a 19th-century military cipher that had remained unsolved for over two centuries, according to a report by TechRadar. The breakthrough demonstrates how modern artificial intelligence systems can rapidly process complex historical codes that previously required weeks of manual human effort.

The decoded message originates from 1809 and details troop positions for Napoleon Bonaparte’s forces and their allies. Japan Today reported that AI engineer Carter Church utilized the GPT-6 Astra model to crack the code. The letter had been sitting ignored in an online archive until Church attempted to decode its symbols, numbers, and letters using the advanced language model.

Context and Background

Church stated that the AI provided persistence that manual decoding lacked, completing the task in just six hours. He expressed satisfaction with the result, which was subsequently verified by Satoshi Tomokiyo. Tomokiyo is a patent engineer who maintains an online list of unsolved ciphers for amateur cryptographers to tackle, confirming the accuracy of the AI's translation.

Simultaneously, Anthropic’s Claude AI model has analyzed cryptographic key generation methods, specifically focusing on elliptic curve seeds. Cyber Security News reported that an internal Claude model traced the origins of five seeds generated by the US National Security Agency in less than 30 minutes. These seeds have been public for more than 25 years.

Who Is Affected

For over two decades, claims persisted that the NSA-generated elliptic curve seeds contained weaknesses or backdoors. The Claude AI analysis determined that no such weaknesses exist and that there are no hidden inside jokes within the data. Anthropic noted that most researchers already trusted these curves, meaning no immediate action is necessary for users relying on this encryption standard.

The AI model also questioned why human cryptographers had not discovered the solution in 25 years, according to internal thinking logs. This interaction highlights the growing capability of AI to outperform humans in core mathematical tasks while providing rapid verification of long-standing cryptographic assumptions.

What should you do?

Security professionals should monitor developments in AI-driven cryptanalysis as these tools become more capable of breaking historical and potentially modern codes. Organizations should continue to rely on established cryptographic standards while staying informed about new verification methods provided by AI models.

What to do and how to stay safe: OpenAI

• Review current encryption protocols to ensure they align with latest industry standards and verified cryptographic seeds.
• Monitor security feeds for updates on AI capabilities in cryptanalysis and code-breaking.
• Verify the integrity of cryptographic keys used in sensitive communications.
• Stay informed about vendor announcements regarding AI-assisted security testing and verification.

Step-by-step guide: Prompt Injection Warning Signs: How to Spot AI Manipulation

General security guidance from the Threat Times newsroom. It is not confirmed advice from the organisations named in this story.

Frequently asked questions

Did the AI find any backdoors in the NSA cryptographic seeds?

No, the Claude AI model determined that there are no weaknesses or backdoors in the five elliptic curve seeds generated by the NSA.

How long did it take GPT-6 Astra to crack the 19th-century cipher?

The AI model completed the decryption of the 1809 military message in just six hours, a task that would have taken weeks manually.

Who verified the accuracy of the decoded historical message?

Patent engineer Satoshi Tomokiyo, who maintains a list of unsolved ciphers for amateur cryptographers, verified the result produced by the AI.