Claude Solves a 25-Year-Old Cryptography Mystery: Were Internet Security Codes Secretly Rigged?

For more than twenty-five years, an important mystery has surrounded some of the mathematical codes used to protect internet communications. These codes help secure banking transactions, digital signatures and other sensitive information. Although the mathematics behind them was publicly available, nobody outside their original designers could explain exactly how certain starting numbers had been chosen.

This lack of transparency created suspicion. Could the numbers have been deliberately selected to contain a hidden weakness, allowing an intelligence agency or some other organisation to break supposedly secure communications?

The mystery concerned five mathematical systems known as elliptic curves, approved by the United States National Institute of Standards and Technology (NIST). These curves form part of the foundations of modern digital security. Their security depends upon mathematical problems that are extraordinarily difficult for ordinary computers to solve.

In October 2026, researchers at the artificial intelligence company Anthropic announced that an experimental version of Claude had solved much of this mystery. Remarkably, the AI reportedly reconstructed the original method of choosing the mysterious numbers in less than half an hour.

The explanation turned out to be surprisingly ordinary. Rather than being derived from some elaborate secret formula, the numbers had apparently been produced using a standard computer program available in the 1990s. The program generated sequences of seemingly random numbers, beginning with a value based upon the date and time.

Claude worked backwards through the mathematical evidence and identified the process that could have produced the published numbers. For one important system, known as P-256, it traced the sequence back to 16 August 1997, at precisely 1:52:34 in the morning, Greenwich Mean Time.

This was rather like finding a mysterious set of lottery numbers and discovering not only the machine that produced them, but also the exact time it was switched on.

Anthropic subsequently published computer code allowing other researchers to reproduce the findings. Independent cryptographers have reportedly confirmed the main reconstruction.

Does this mean that internet security has been broken? Fortunately, no. Claude did not discover a method of stealing passwords, breaking encrypted communications or forging digital signatures. It discovered how certain publicly known mathematical parameters had originally been generated.

Indeed, the finding may actually strengthen confidence in these security systems. The suspicion had been that the numbers might have been secretly manipulated to introduce a hidden weakness. Now that their origins can largely be reconstructed using an ordinary computer program, the possibility of such deliberate manipulation appears considerably less plausible.

There are still some unanswered questions about why particular numbers were selected or rejected. Nevertheless, a mystery that had persisted for a quarter of a century has been substantially resolved.

The wider significance lies in what this episode tells us about artificial intelligence. Claude did not merely retrieve an answer from an existing document. It examined a complicated mathematical puzzle, recognised clues pointing towards old computer software, and reconstructed a process that human investigators had been unable to establish conclusively for decades.

This is an impressive demonstration of AI as a research tool. Problems that once required months or years of specialist investigation may increasingly be solved in minutes by systems capable of exploring enormous numbers of possibilities and recognising patterns that human researchers overlook.

There is, however, a more unsettling side to this achievement. The same capabilities that allow artificial intelligence to uncover forgotten mathematical secrets might also help identify weaknesses in software, financial systems and digital infrastructure. The present discovery was reassuring because it did not expose a security vulnerability. Future discoveries may not always be so benign.

It would be a mistake to conclude that AI has suddenly defeated modern cryptography. The underlying mathematical problems protecting these systems remain extremely difficult. But the episode demonstrates how rapidly the boundary between what computers can and cannot discover is changing.

A mystery that had survived since the 1990s was reportedly solved by an AI system in less time than it takes to drink a cup of coffee. That is an extraordinary achievement, and another indication that artificial intelligence is becoming a powerful instrument for uncovering knowledge that has long remained beyond human reach.