For 167 years the Riemann hypothesis has been the Everest of mathematics. On Friday, a mathematician at Anthropic said he and Claude had reached the summit, and within hours a rival, all-human claim appeared too.
Levent Alpöge, a mathematician employed by AI company Anthropic, said on Friday, November 20, that he has found a proof of the Riemann hypothesis with the help of the company’s Claude AI model. “I believe I have, working with Claude, discovered a proof of the Riemann Hypothesis,” he wrote in a statement on his personal website, as reported by Scientific American. He went further, saying the method also resolves the much broader Generalized Riemann Hypothesis. The claim has not yet been formally verified, but early independent checks of the computer-checked proof have come back positive.
What exactly was claimed
Alpöge did not release a traditional paper. Instead, he posted the proof written in Lean, a programming language that mathematicians use to check every logical step of an argument by machine. He told readers it would be “obviously inappropriate to attempt to post some massive unreadable travesty of ‘slop’,” and said an unpublished human-language version ran to more than 250,000 pages, according to New Scientist.
In his statement he also says he and Claude “formalized the arguments without assumptions,” and that he expects the method can eventually stretch to the Extended Riemann Hypothesis as well. Anthropic itself has not commented on or involved itself in the announcement, The Verge reported. Alpöge wrote that he texted CEO Dario Amodei about the result “late Sunday night” and suggested he held off publishing until after the company’s stock market debut on Thursday, November 19. “The precise timing is horrible,” he wrote.
Why this problem matters so much
Proposed by German mathematician Bernhard Riemann in 1859, the hypothesis concerns the zeros of a function now known as the Riemann zeta function, and through them the way prime numbers are spread along the number line. Primes look random at first glance; the hypothesis says their distribution follows a deep and precise order. It is one of the seven Millennium Prize Problems set by the Clay Mathematics Institute, each carrying a $1 million reward. Only one of those, the Poincaré conjecture, has been officially solved.
A huge body of mathematics has been built on the assumption that the hypothesis, and its generalized form, are true. “There are numerous results that are conditional on the generalised Riemann hypothesis, and now those are theorems!” James McKee of Royal Holloway University of London told New Scientist. Hector Pasten of the Pontifical Catholic University of Chile told Scientific American that a correct, understood proof would act “something like a Rosetta Stone” between branches of mathematics.
How the verification is going
A Lean proof that compiles without errors is usually treated as strong confirmation, but two questions remain: whether the code runs for other people, and whether the theorem stated in the code really is the Riemann hypothesis rather than something subtly different. On the first point, independent researchers including MIT mathematician Andrew Sutherland have compiled the code themselves. “The Lean proof has been compiled and independently re-checked by the Lean kernel with only the three standard axioms,” Sutherland said, in a follow-up report from Scientific American. “Honestly the chance that anything is fishy here is approximately zero.”
The second check is still pending. Kevin Buzzard of Imperial College London said experts were awaiting a successful “comparator run,” which confirms that the code does not quietly redefine the symbols in the statement and that it compiles under several independent checkers, guarding against an unknown bug in Lean itself. Scientific American said such a run typically takes a matter of hours. Even then, the Clay institute’s rules require a solution to be published and survive at least two years of scrutiny, with general acceptance among mathematicians, before any prize can be considered.
A rival, human-only claim
The story took a twist within hours. Ravi Vakil, president of the American Mathematical Society and a professor at Stanford, posted a draft proof of the Generalized Riemann Hypothesis by Stanford postdoc Alessandro Masullo, stressing that “it is all human, with no AI,” TechCrunch reported. Vakil said the plan had been to spend a year polishing the writing, but “now it seems best to put out the current draft.” Masullo says the work is the product of 12 years of effort.
That claim is at a much earlier stage. Stanford mathematician Brian Conrad, who has been giving Masullo feedback, told Scientific American that the paper is “undergoing major revision” and needs to reach a more final form before its success can be judged. Scientific American added that some mathematicians are already skeptical of it.
What it means for AI and mathematics
The announcement caps an extraordinary year for AI in mathematics. In July, Alpöge used Claude to show that the decades-old Jacobian conjecture is false, offering a tiny counterexample. In September, OpenAI claimed a solution to the Navier-Stokes Millennium problem, a result that is considered credible but was clouded by allegations, which OpenAI denies, that it drew on unpublished work by mathematicians. Anthropic said in August that it did not expect the techniques Claude had used on the Riemann hypothesis at the time to lead to a full proof.
In our view, the most striking part is not the AI hype but the method: a result that may be correct beyond reasonable doubt yet is not yet understood by any human. Even Alpöge has committed to producing a readable explanation. Sutherland called it “a human triumph” but added that “there is much work to be done to understand the mathematical details of the proof.” McKee summed up the mood: “It is both a wonderful time and a terrible time to be a research mathematician.”
What happens next
The immediate milestone is the comparator run on Alpöge’s code. After that, mathematicians will need to turn hundreds of thousands of lines of machine-checked logic into an argument people can follow, while Masullo’s human draft goes through its own revision and review. Neither claim has been accepted by the Clay Mathematics Institute. For now, the most famous open problem in mathematics may have been solved, twice, in a single day, but the verdict still belongs to the experts.