22 November 2026, Sunday, 11:41
Support
the website
Sim Sim,
Charter 97!
Categories

Claude Proved Riemann's Hypothesis, Which Mathematicians Had Been Struggling With For 167 Years

Claude Proved Riemann's Hypothesis, Which Mathematicians Had Been Struggling With For 167 Years

The written version of the evidence exceeded 250,000 pages.

A problem that mathematicians had been unable to solve for 167 years—and for which a $1 million prize is being offered—may have been solved by AI. Levent Alpoge , a mathematician at Anthropic , stated that, together with Claude, he proved the Riemann Hypothesis. It is not specified which version of the model he used. Previously, Anthropic has utilized unreleased research versions—in which dozens of AI agents work collaboratively—to make mathematical discoveries. These are not available to regular users, according to itc.ua.

What Is the Riemann Hypothesis

Prime numbers (2, 3, 5, 7, 11…) are the “atoms” of arithmetic, but at first glance, they appear to occur randomly. In 1859, Bernhard Riemann proposed that their distribution is related to the zeros of the so-called zeta function, and that all significant zeros lie on a single vertical line. Computers have already checked more than 10 trillion such zeros—and all of them match the prediction. But no number of examples proves that this will hold true for an infinite set of numbers. Therefore, the hypothesis is one of the Clay Institute’s seven “Millennium Problems.”

Many theorems in mathematics begin with the words “Suppose the Riemann Hypothesis is true.” If the proofs are confirmed, all these results will gain a solid foundation.

How Claude Helped

“I believe that, while working with Claude, I discovered a proof of the Riemann Hypothesis,” Alpoge wrote in his four-page statement.

According to him, the method also applies to the generalized Riemann Hypothesis. Key breakthroughs occurred in mid-November, and on November 16, they succeeded in arriving at a generalized result.

Alpoge says he led the research, but in the final stages, Claude “significantly outpaced” him. Due to the coincidence with Anthropic’s IPO, the mathematician acknowledged that the “timing was terrible,” but called the cynical explanation erroneous.

Why mathematicians aren’t rushing to celebrate

The written version of the proof, generated by a computer, exceeded 250,000 pages. Therefore, Alpoge published it only as code for the Lean proof-checking system—17 million lines in 14 archives.

“Let’s first see if it compiles,” said Kevin Bazzard of Imperial College London.

Thomas Bloom of the University of Manchester called this, “if true,” the greatest achievement in AI, but urged skepticism: the AI could have exploited a hidden error in the code.

In other words, this is still just a claim, not a recognized result: according to the Clay Institute’s rules, the prize is not considered until at least two years after publication. On the same day, mathematician Alessandro Masullo presented his own proof of a generalized version without AI—a 940-page document that has not yet been verified either. In October, OpenAI had already published an AI-generated proof of a weaker version of the hypothesis.

So, AI—which is already finding vulnerabilities on a massive scale and, in the form of Gemini, has disproved a 59-year-old hypothesis about nuclear fusion—is now claiming the title of author of the greatest discovery in mathematics. Alexey Skorobogatov of Imperial College compared this to the invention of the nuclear bomb.

“What seemed impossible a year ago has become a reality,” says Rainer Dietmann, fearing that it will become increasingly difficult for people to even understand such proofs.

Write your comment

Follow Charter97.org social media accounts

▲