
The most famous unsolved problem in mathematics may have just met its match: artificial intelligence.
For 167 years, the Riemann Hypothesis has baffled the world’s greatest mathematicians. It concerns a mysterious pattern hidden in prime numbers, and solving it carries a $1 million prize.
Now, a mathematician at AI company Anthropic claims he has finally cracked it, with help from Claude, the company’s AI chatbot.
“I believe I have, working with Claude, discovered a proof of the Riemann Hypothesis,” Levent Alpöge wrote in a four-page statement published on his website.
Alpöge did not disclose which version of Claude he used. However, Anthropic has previously used unreleased research versions of its AI for mathematical discoveries, including earlier work on the Riemann hypothesis. These systems operate with dozens of AI agents working together and are not available to ordinary Claude users.
The claim has not yet been independently verified. But if it holds up, this would be one of the biggest mathematical breakthroughs in generations. It could settle countless questions about numbers and mark a turning point in AI’s ability to make major scientific discoveries.
The 167-Year-Old Mystery of Prime Numbers
To understand why mathematicians are so excited, we need to talk about prime numbers for a minute.
Primes are numbers greater than one that can be divided evenly only by themselves and one: 2, 3, 5, 7, 11, 13, and so on.
They are the building blocks of arithmetic. Every whole number greater than one is either prime or can be broken down into a unique combination of primes.
But primes aren’t predictable at all.
Sometimes they appear almost side by side, like 17 and 19. Elsewhere, enormous gaps separate them. They become increasingly rare as numbers grow larger, but their exact spacing seems unpredictable.
Mathematicians have long known how to estimate how many primes appear below a given number. What they haven’t established is precisely how far the actual count can stray from that estimate.
In 1859, German mathematician Bernhard Riemann proposed an astonishing connection between primes and a mathematical object called the Riemann zeta function.
This function takes numbers as inputs and produces other numbers as outputs. At certain inputs, its output becomes exactly zero. Mathematicians call these special inputs its zeros.
Riemann conjectured that all the important, non-obvious zeros lie on a single vertical line, at a position mathematicians call one-half.

All this may sound a bit confusing and perhaps even esoteric for the rest of us not versed in high-order mathematics, but the main takeaway is that the location of those zeros determines how accurately mathematicians can describe the distribution of primes.
If every relevant zero lies on that line, the fluctuations in the number of primes follow much tighter mathematical limits than we can currently prove.
According to the Clay Mathematics Institute, researchers have checked the first 10 trillion relevant zeros. Every one satisfies Riemann’s prediction.
But even checking a trillion trillion examples cannot prove that something holds for infinitely many numbers. Mathematics can be very strict like that.
That’s why the Riemann hypothesis remains one of the seven Millennium Prize Problems, each carrying a $1 million reward. Despite the generous cash prize, some of the best mathematicians in the world have devoted very little time to this problem, finding it futile and not worth their time.
The generalized version goes further, extending the same fundamental prediction to a much larger family of mathematical functions. Proving it would settle numerous results that mathematicians have so far established only by assuming the conjecture is true. Alpöge claims he and Claude have actually solved this generalized version of the Riemann Hypothesis.
How Claude Apparently Helped Crack the Problem
Levent Alpöge in 2024. Credit: Wikimedia CommonsAlpöge explains that he had wanted to tackle the Riemann hypothesis since his student days and spent years learning mathematical techniques that might eventually prove useful.
Working with Claude, he began combining ideas from different branches of mathematics.
One crucial inspiration came from a related version of the Riemann hypothesis that mathematicians had already solved in a different mathematical setting.
The challenge was adapting those techniques to ordinary whole numbers. According to Alpöge, Claude helped find a workaround involving repeated transformations called theta lifts. Roughly speaking, these transformations would recreate a property that the existing method needed but couldn’t directly obtain.
Alpöge describes the mathematical argument as extraordinarily complicated, drawing together ideas from across number theory.
He says he guided the research toward the right approach, but Claude eventually surpassed his own technical understanding.
“No, Claude flew, and danced, and sparkled near that finish line,” he wrote.
He says the crucial breakthroughs came in mid-November, with the claimed generalized result reached on November 16.
But Alpöge hasn’t published a conventional mathematical paper explaining the complete proof.
He says a computer-generated written version exceeded 250,000 pages, and he expects the argument to become considerably shorter once mathematicians understand it better.
Instead, he released the proof as code written in a mathematical programming language called Lean.
And that’s where things get particularly interesting.
Can a Computer Really Verify the Proof?
Lean is a specialized system designed to check mathematical reasoning with extraordinary rigor.
Imagine a mathematician presenting a proof consisting of thousands of steps. A human reviewer must examine those steps and determine whether each conclusion follows logically from what came before.
Lean performs that logical checking mechanically and extremely fast at the speed of a computer. If a proof passes all its checks, mathematicians can be extremely confident that the reasoning is sound.
Alpöge has released his proposed proof in 14 downloadable archive parts.
But that doesn’t mean everything is solved and done.
Lean can verify the logic of a mathematical statement without guaranteeing that the statement itself is exactly the one its author intended to prove.
Before the Riemann Hypothesis is declared officially solved, other mathematicians must confirm that the code accurately defines the Riemann hypothesis, that it doesn’t rely on hidden assumptions, and that the entire project successfully compiles.
“Let’s see if it compiles first,” Kevin Buzzard, a mathematician at Imperial College London, told New Scientist.
Until then, the million-dollar prize is on hold. Under Clay’s rules, a solution must appear in a qualifying publication and achieve broad acceptance among mathematicians, with at least two years passing before consideration.
Meanwhile, the extraordinary claim arrives amid an extraordinary acceleration in AI-assisted mathematics.
In August, Anthropic reported that Claude had improved a longstanding result concerning the proportion of zeta-function zeros known to lie on Riemann’s predicted line, raising the established lower bound from 41.6% to 67.2%.
That was significant progress, but nowhere near a complete proof.
The new claim would represent an entirely different level of achievement.
James McKee, a mathematician at Royal Holloway University of London, said “It is both a wonderful time and a terrible time to be a research mathematician.”
Alpöge himself acknowledges that resolving the problem could feel like losing something precious. For generations, the Riemann hypothesis has been a guiding challenge, inspiring entire careers.
But he sees a much bigger beginning.
“It is natural to feel that this is an end, in my case a loss of a principal guiding light,” he wrote. “But what I saw during the course of progress so amazed me that I believe it is one dawn across science of something far greater.”
