AI keeps breaking old encryption mysteries — and is dunking on humans too

AI models are proving to be formidable code-breaking machines and are challenging human capabilities in the realm of encryption. Recently, it has been reported that an AI engineer, Carter Church, used OpenAI's GPT-6 Astra model to crack a 19th-century military cipher that had eluded experts for over two centuries. The message, originating from Napoleon Bonaparte's era, was initially overlooked in an online archive but was swiftly decoded by the AI in just six hours.

The second instance of AI demonstrating its prowess in code-breaking involves cryptographic key generation, specifically with elliptic curve encryption. An Anthropic's internal Claude AI model has traced the origins of five elliptic curve seeds generated by the US National Security Agency (NSA) in less than 30 minutes, debunking claims that these seeds contained weaknesses that could be exploited as backdoors.

While most researchers trusted the curves already, Claude's findings provided a fresh perspective on the matter, showcasing its ability to outperform humans in the core mathematics and humorously remarking on the obviousness of the discovery.

These breakthroughs in AI capabilities highlight the increasing prowess of AI models in various tasks, including code-breaking and testing widely used encryption security protocols. As AI continues to advance, it is clear that these models are not only breaking old encryption mysteries but are also "dunking on humans" in the process.