Mathematics had a day it won't forget on Friday. In the space of a few hours, two separate teams claimed to have solved the Riemann hypothesis — the most famous unsolved problem in maths, untouched for more than 160 years. One proof was built with an artificial intelligence. The other was written entirely by a human. And, remarkably, independent experts say the AI version is very probably right.
What Claude and its creator say they found
Levent Alpöge, a number theorist who works at Anthropic, announced on Friday that he and the company's AI model Claude had found a proof of the Riemann hypothesis. According to his announcement, the same method also settles the even broader Generalised Riemann Hypothesis.
Rather than publishing a traditional paper, Alpöge released the argument as computer code: a 14-part archive containing more than 17 million lines written in Lean, a language designed so software can check a proof line by line. He has been working with Claude on the problem since early October, and several of the key steps are dated 15 November.
Alpöge has been strikingly honest about the process. He says he guided Claude through weeks of investigation before the model pulled far ahead of him near the finish line — and he has acknowledged that he does not yet fully understand every argument the AI used to produce the result.
Why experts are taking it seriously
The reason mathematicians are excited, rather than dismissive, is that the code has now been independently checked. Andrew Sutherland, a mathematician at the Massachusetts Institute of Technology, separately compiled Alpöge's result in Lean and had it rechecked by the Lean kernel. "Honestly the chance that anything is fishy here is approximately zero," he told Scientific American. According to The Hindu, two separate proof-checking systems have also accepted the result.
That is a long way from a done deal, though. Researchers are still running a final layer of checks, known as a comparator run, to make sure the code does not quietly redefine the meaning of the hypothesis itself. And the Clay Mathematics Institute, which offers a $1 million prize for a solution, still lists the problem as unsolved. Under its rules, a proof must be published in a qualifying journal, win general acceptance over two years, and survive scrutiny by a specialist committee — so the prize, if it comes at all, is years away.
The human rival, rushed out the same day
Hours after Alpöge's announcement, a second claim appeared. Ravi Vakil, president of the American Mathematical Society, posted a 940-page manuscript by Alessandro Masullo, a mathematician who has spent 12 years on the problem. Vakil described it in five words: "all human, with no AI."
Masullo's draft had been under private review for months by the Stanford number theorist Brian Conrad, and the plan had been to spend another year polishing it before announcing anything. According to Conrad, Alpöge's announcement prompted them to release the unfinished draft instead.
That claim looks shakier. Masullo has already acknowledged mistakes in his manuscript, and within hours several mathematicians reported problems with it — some of them spotted with the help of AI models. Conrad has confirmed that Vakil's account is accurate, but he says more work is needed and the paper is still undergoing major revision.
Why this problem matters so much
The Riemann hypothesis, first proposed by Bernhard Riemann in 1859, concerns the distribution of prime numbers — the numbers divisible only by 1 and themselves, which form the building blocks of all whole numbers. It predicts that all the non-trivial zeroes of a function called the zeta function have a real part of exactly one half. Proving it would turn a huge number of results that currently rest on assuming it is true into proper theorems.
Contrary to some popular claims, a proof would not reveal the location of every prime number or break modern encryption. Its value is mathematical: after 167 years as a guiding light for number theory, the problem may finally have an answer — twice over, in a single day.
This is a developing story: the final machine checks on the AI-assisted proof are still running, and both claims face intense scrutiny from mathematicians around the world.
