Technology

Anthropic Mathematician Says Claude Helped Him Prove the Riemann Hypothesis

Martin HollowayPublished 3m ago4 min readBased on 1 source
Reading level
Anthropic Mathematician Says Claude Helped Him Prove the Riemann Hypothesis
source:anthropic.com

Levent Alpöge, a mathematician at Anthropic, says he has solved the Riemann hypothesis with the help of Claude, the company's AI model. According to The Verge, Alpöge states that working with Claude he discovered a proof of the Riemann Hypothesis that extends to the Generalized Riemann Hypothesis. The claim has yet to be verified independently.

Alpöge's solution is written in Lean, a proof assistant and programming language that checks mathematical arguments formally, step by step. It has been published only on his website. No journal publication or peer-reviewed version has been reported.

The Riemann hypothesis, first proposed in 1859, concerns the distribution of the nontrivial zeros of the Riemann zeta function. Those zeros carry deep information about how prime numbers are spread along the number line. The problem is one of seven Millennium Prize problems. The Clay Mathematics Institute administers the prize and offers a $1 million bounty for each problem solved. Its rules require that a proposed solution be published, withstand at least two years of scrutiny, and receive general acceptance in the global mathematics community. Only one Millennium Prize problem has been officially solved, the Poincaré conjecture.

Alpöge said he texted Anthropic CEO Dario Amodei about the solution late Sunday night. He also suggested he held off publication until after Anthropic's IPO on Thursday.

Anthropic's own position on the problem has shifted over the past months. The company said in August that it was working on the Riemann Hypothesis with Claude. It had previously stated that it did not expect the techniques Claude used would lead to a proof.

The announcement also arrives in a crowded field of AI-related claims on the Millennium problems. OpenAI claimed in September to have cracked the Navier-Stokes problem, another Millennium Prize problem. Alpöge had himself been working on Navier-Stokes with Tristan Buckmaster, a mathematics professor at NYU.

In my view, the question that matters most here is not who announced first but how the claim will be tested. A Lean proof can be checked mechanically, line by line, but a formal verification only establishes that the argument is valid against the statement as written. Whether the formalized statement matches the conjecture, and whether the community accepts the argument as a resolution, are judgments mathematicians have to make. The Clay rules deliberately place that judgment over a long period rather than at the moment of publication.

The broader context here is a shift in how mathematical claims are produced and checked. Lean gives reviewers a machine-readable artifact to interrogate, which makes the claim more open to examination than a conventional preprint. If the proof holds, the episode will be a concrete test of AI-assisted research on one of the hardest open problems in the field. If it does not, the Lean artifact still gives reviewers a precise record of where the argument fails.