Researchers said on Friday that artificial intelligence (AI) had helped them crack two problems that have defeated mathematicians for generations: the Riemann hypothesis, about prime numbers, and a central question about the Navier-Stokes equations, which describe how fluids move.

Mathematicians are divided over what such proofs are worth when a computer has checked them but no person has read them in full.
Both problems are among the seven Millennium Prize Problems chosen in 2000 by the Clay Mathematics Institute, a US foundation that funds mathematical research. It offers $1 million for a solution to each. Specialists in computer-checked mathematics have certified the Riemann proof. The Navier-Stokes claim has not yet been checked by anyone outside the team that made it.
Riemann hypothesis
Prime numbers are whole numbers greater than one that can be divided evenly only by themselves and one, such as 2, 3, 5, 7 and 11. They thin out as numbers get bigger, and they turn up at irregular intervals.
In 1896, mathematicians proved the prime number theorem, a formula that estimates how many primes lie below any given number. The estimate is right on average, but the real count wanders above and below it. Mathematicians wanted to know how far it can wander.
{{/usCountry}}In 1896, mathematicians proved the prime number theorem, a formula that estimates how many primes lie below any given number. The estimate is right on average, but the real count wanders above and below it. Mathematicians wanted to know how far it can wander.
{{/usCountry}}German mathematician Bernhard Riemann tied that question in 1859 to a formula now called the Riemann zeta function. A function takes in a number and returns another number. The inputs for which the zeta function returns zero are called its zeroes. Riemann showed that their positions decide how far the count of primes strays from the estimate.
These inputs are complex numbers, which have two parts and can be pictured as points on a flat map called the complex plane. Some zeroes, known as trivial zeroes, fall at -2, -4, -6 and so on. Riemann's hypothesis is that all the others, the non-trivial zeroes, lie on one vertical line on that map, where the first part of each number, called its real part, is exactly one-half.
The Clay institute says on its website that the prime number theorem gives the average spread of the primes, while the Riemann hypothesis tells mathematicians about the deviation from that average. If the hypothesis is true, the primes stray from the estimate as little as they possibly can. Many published results in mathematics start by assuming it is true. A proof would put all of them on solid ground.
Unsolved
There are infinitely many non-trivial zeroes. Using computers, mathematicians Dave Platt and Tim Trudgian confirmed in a 2021 paper that every zero up to a very large height, which amounts to trillions of them, lies on the line. Checking examples can never cover an infinite set, though. A zero off the line would make the hypothesis false.
Mathematicians have managed partial results. British mathematician GH Hardy proved in 1914 that infinitely many zeroes lie on the line, which still left room for others elsewhere. German mathematician David Hilbert put the problem on his list of 23 major open questions in 1900, a century before Clay attached a prize to it.
Also read: Scientifically Speaking: As AI cracks hard Math, how will my son’s generation thrive?
The Navier-Stokes problem
The Navier-Stokes equations, worked out in the 19th century by French engineer Claude-Louis Navier and Irish physicist George Gabriel Stokes, describe how liquids and gases move. They are used to design aircraft and to forecast the weather.
The prize question concerns an idealised fluid that cannot be squeezed into a smaller volume, known as an incompressible fluid, moving in three dimensions. Mathematicians want to know whether such a fluid, if it starts out moving smoothly, with its speed changing gradually from point to point, always stays smooth.
The alternative is a blowup, also called a singularity, in which the fluid's speed in the equations shoots up to infinity within a finite time. Real water would never move infinitely fast. A blowup would mean the equations had stopped giving a complete description of the flow, which OpenAI's account of the problem calls a breakdown in how the equations model the fluid.
Unsolved too
The equations are nonlinear, meaning the fluid's motion feeds back on itself. Small swirls can grow stronger as a result.
French mathematician Jean Leray showed in 1934 that rough, so-called weak solutions to the equations always exist. Whether those solutions stay smooth has been an open question ever since, which is why OpenAI described the problem as roughly 90 years old.
The same question in two dimensions was settled decades ago. In three dimensions, the main quantity mathematicians can keep in check is the fluid's total energy. That control is too weak to rule out the flow squeezing into ever smaller, faster whirls.
Clay's official statement of the problem, written by Princeton University mathematician Charles Fefferman, offers four routes to the prize. Two involve proving that smooth flows last forever. The other two involve showing they can break down. In those two, an outside force, such as stirring, is allowed to push the fluid.
The prize
Clay will consider a solution only after it has been in print in a qualifying publication for at least two years and has won general acceptance among mathematicians, according to the institute's rules. Only one Millennium problem, the Poincaré conjecture, has been solved.
What researchers say
Levent Alpöge, a mathematician on the technical staff of AI company Anthropic, posted a four-page note on his personal website on November 20. "I believe I have, working with Claude, discovered a proof of the Riemann Hypothesis," he wrote. Claude is the AI model made by Anthropic. Alpöge said the same method also proves a broader version, the generalised Riemann hypothesis.
He did not publish a conventional paper. He released the proof as 17 million lines of code in Lean, a programming language for writing out proofs so that a computer can check each step follows from the ones before, the Associated Press (AP) reported. Alpöge said his work with the AI produced more than 250,000 pages, according to AP. That written version has not been released.
A Lean check proves only the exact statement typed into the computer, so experts first had to confirm that the code states the Riemann hypothesis correctly. The Lean Focused Research Organization, which develops Lean, and the Mathlib Initiative, which supports Lean's shared library of mathematics, examined the statement and the proof. "We hereby certify that both the statement and the proof of the Riemann Hypothesis are correct," they said in a joint statement on November 21.
OpenAI made the first Navier-Stokes claim. On September 8, 2026, the company said about 10,000 AI agents running an internal model had produced a proof, with a Lean version, that a fluid starting at rest can blow up in a finite time. The construction relied on an outside force, which Clay's problem statement allows. That left open the version with no outside force. OpenAI said: "We do not intend to claim the Millennium Prize for this result." Clay said on September 11 that the problem had "apparently been settled", adding that its review would be "deliberately unhurried".
OpenAI's announcement came hours after Tristan Buckmaster, a New York University mathematician who studies fluid equations, and Alpöge released AI-assisted results on related equations. Buckmaster questioned whether OpenAI had drawn on their work, including drafts the pair had kept in Codex, OpenAI's AI coding tool. OpenAI said its researchers and agents "did not see any of their work through any means until they released it publicly". It said an internal investigation had confirmed that Buckmaster's Codex prompts "could not have influenced the system in any way, including through training".
On November 20, Buckmaster and Alpöge posted a 79-page paper claiming a blowup with no outside force at all. A file of numerical data needed to check a key step was not released with it, according to the project's online repository. The paper says the file will follow. Until then, the result cannot be fully checked.
Also read: OpenAI claims to have solved hundreds of long-standing maths problems
What it means for mathematics
For centuries, mathematicians have accepted a proof after experts read it and agreed it was sound. In the Riemann case, a computer has checked the argument, but a person has not yet read all of it.
Alex Kontorovich, a Rutgers University mathematician who leads a project to write the proof of the prime number theorem in Lean, told AP that "the computer is extraordinarily picky, way pickier than I am". He said he had not yet fully understood the proof. "But I have to do the work," he said.
Martin Bridson, president of the Clay institute and an Oxford University mathematician, called it "an announcement of historic significance". "But the journey from today's announcement to a place of real understanding will be long and difficult," he told AP. Hours after the announcement, Clay's website still listed the Riemann hypothesis as unsolved, AP reported.
The researchers also describe the AI's part in different ways. Alpöge wrote that "Claude far outpaced me near the finish line", according to AP. Buckmaster has given more weight to earlier human work. In a statement in September, he said the credit for the basic idea behind his fluid work with Alpöge "goes to Diego Córdoba and Luis Martínez-Zoroa", two mathematicians who had spent several years on the approach. He and Alpöge used large language models "to push their program to completion", he wrote. Large language models, or LLMs, are AI systems such as Claude that learn from huge amounts of text.
Many leading mathematicians are uneasy about how fast AI-made proofs are appearing. On September 11, 2026, 25 winners of the Fields Medal, the highest award in mathematics, signed a statement saying the goals of AI companies and of mathematics are "severely misaligned". The signatories included Terence Tao of the University of California, Los Angeles, and James Maynard of Oxford. They argued that AI companies treat famous problems as tests for their models, while mathematicians value the understanding gained by solving them.
Earlier, in June, a group of researchers from 15 universities published the Leiden Declaration, which the International Mathematical Union has endorsed. It warned of AI proofs that look convincing but contain errors, and of breakthroughs announced before proper scrutiny.
Both claims were also made during a race between AI companies. Buckmaster wrote on November 20: "After last time, we couldn't risk waiting, so the Navier–Stokes write-up is unpolished and heavily LLM-assisted." Alpöge's announcement followed Anthropic's stock market debut, which raised a record $81 billion, AP reported. Anthropic did not respond to AP's request for comment on Alpöge's claim.
For now, the Riemann proof has passed a machine check, but mathematicians have yet to understand it. Clay's rules require at least two years in print before the institute weighs any prize. Outside experts cannot test the unforced Navier-Stokes claim until the missing data file is released.
ABOUT THE AUTHOR
HT Explainers Desk delivers in-depth explainers and analysis on the stories shaping India and the world, from SimplyPut, Hindustan Times’ dedicated explainer section. The desk unpacks the science, policy, geopolitics and market forces behind major news events, breaking down complex developments into clear, accessible reporting for readers who want more than the headline. Explainers draw on the expertise of the journalists covering the story and insight from leading subject experts, translating specialist detail into plain language without losing precision or nuance. Coverage spans national and global news — from health and science breakthroughs to trade, technology, cybersecurity, aviation, energy and public policy — going beyond what happened to explain how it works, why it matters and what to watch next. Written for readers who want context, and not just the headlines, HT Explainers Desk brings together original reporting, data and expert analysis to make sense of the news as it develops. It is part of Hindustan Times, India's leading source for news, in-depth analysis and explanatory journalism since 1924.