A four-page note on a mathematician’s personal website is making an extraordinary claim: Claude helped prove the Riemann Hypothesis, a problem that has resisted mathematicians since 1859.
Levent Alpöge, a number theorist who works at Anthropic, posted the claim on November 20, along with files containing a proposed proof in Lean, software designed to check mathematical reasoning.
“I believe I have, working with Claude, discovered a proof of the Riemann Hypothesis,” he writes.
That’s a sentence worth stopping for.
Scientific American is reporting the claim, too. Its coverage adds weight to the argument that this should be taken seriously. Between Alpöge’s credentials, the released formalization, and attention from an established science publication, there’s enough here to justify some excitement.
Independent verification of the proof is still pending. I’m particularly interested in what happens when other mathematicians open those files.
Anthropic mathematician Levent Alpöge says he found a proof of the Riemann Hypothesis with help from Claude. If the proof holds up, it would solve one of math’s biggest open problems.
- Corey Noles, The Neuron.
Why the Riemann Hypothesis matters
The Riemann Hypothesis concerns the distribution of prime numbers: numbers like 2, 3, 5, and 7 that can’t be divided evenly by anything except themselves and 1. Primes become less common as numbers grow, but their precise distribution contains irregularities that mathematicians have spent generations trying to understand.
The hypothesis describes a deep constraint on those irregularities. Technically, it says that all the “nontrivial” zeros of the Riemann zeta function lie on a particular vertical line in the complex plane, with real part equal to one-half.
You don’t need to follow the complex analysis to appreciate the stakes. A proof would establish much tighter control over how primes are distributed. It would also remove an assumption from mathematical results that currently depend on the hypothesis being true.
Alpöge’s claim goes further. His note says the method also resolves the Generalized Riemann Hypothesis, which extends the idea to a family of related mathematical functions. He expects another extension, the Extended Riemann Hypothesis, to follow with more work. That last part remains an expectation.
There’s useful context for why this claim deserves attention now. In August, Anthropic reported that an experimental version of Claude had improved a longstanding result concerning the zeta function, raising the proven lower bound on the proportion of zeros lying on the critical line from 41.6% to 67.2%.
That wasn’t the Riemann Hypothesis being “two-thirds solved.” The full hypothesis concerns every relevant zero, and closing the remaining gap requires a proof that covers them all. Still, Anthropic said mathematicians had examined the earlier result and released both a paper and a formalization.
The new claim arrives against that background of substantive work on the same mathematical territory.
What Alpöge says Claude contributed
Alpöge describes a collaboration in which he directed the effort toward a promising approach, and Claude supplied major advances as the argument developed. He credits the model with pulling ahead near the finish.
His sketch draws on a technique used in another mathematical setting where an analogue of the Riemann Hypothesis is already known. The difficulty is carrying the useful idea across to ordinary integers. According to the note, Claude helped find a workaround involving a long chain of transformations called theta lifts.
Specialists will have to assess that explanation. For everyone else, the important detail is that Alpöge attributes mathematical ideas to Claude, including a central step in the proposed argument.
The unusual part is how the proof has been released. There’s no conventional paper explaining the complete argument yet. Alpöge says it’s extremely long and has instead published the Lean formalization across 14 archive parts.
How researchers will check the proof
Lean gives this announcement a concrete route to scrutiny. A proof written in its formal language can be checked against explicit definitions and logical rules. Researchers can inspect what the final theorem actually says and which assumptions it depends on.
Those details matter. A computer can correctly verify a statement that falls short of the problem people intended to solve. Lean also supports standard foundational axioms, so the meaningful check is whether the result relies on acceptable foundations without additional assumptions or proof gaps. Its documentation explains how those dependencies work.
Kevin Buzzard, a mathematician at Imperial College London known for his work with Lean, offered a suitably practical response to New Scientist: “Let’s see if it compiles first.”
A successful independent check would be enormously consequential. Researchers would need to establish that the files build, that the theorem faithfully states the Riemann Hypothesis, and that its dependencies hold up. The generalized claim needs the same scrutiny.
Then comes the work of explaining the mathematics.
A checked proof still needs an explanation
Alpöge is candid about not having mastered all the territory involved in the argument. He intends to understand it more fully, communicate it, and produce a proper written account.
That admission makes this especially interesting. If the proof survives, mathematicians could have strong formal evidence for a result while still working to understand the ideas that make it true. Simplifying the argument and identifying which techniques can help elsewhere would become substantial research projects in their own right.
As The Neuron has covered in earlier AI research announcements, verification and explanation are becoming increasingly prominent parts of the scientific work around model-generated results.
For researchers, understanding how an argument works makes it possible to teach the ideas, adapt them, and recognize connections to other problems. A verified answer to a famous question would be extraordinary. The mathematics inside that answer could give people years of further work.
What happens next
The million-dollar prize would follow a much slower process. The Clay Mathematics Institute’s rules require publication in a qualifying outlet, at least two years after publication, and general acceptance by the global mathematics community before it will consider a proposed solution.
For now, the useful thing to watch is independent examination of the released proof.
If it holds, this would be a remarkable example of a mathematician working with AI to reach a result that generations of researchers couldn’t. It would give us a specific, inspectable achievement to study, including how the human direction and the model’s contributions fit together.
I’m comfortable being excited about that possibility. The next report I want to read is from researchers who’ve checked the files and can begin explaining what Claude and Alpöge found.
Editor’s note: This article originally appeared on our sister publication, The Neuron.


