A mathematician working at Anthropic says he has used the company's AI model, Claude, to prove the Riemann hypothesis, a problem many experts regard as the single most important unsolved question in mathematics.
Levent Alpöge made the claim on November 20th, one day after Anthropic completed what was described as the largest initial public offering in history, according to a report by Livemint. The proof has not yet been reviewed by other mathematicians, but its arrival has already drawn serious attention from leading figures in the field.
James Maynard, a mathematician at the University of Oxford, put the stakes plainly: "If it fully checks out, it is clearly a historic milestone."
A Problem That Has Stood for More Than a Century
The hypothesis is named after Bernhard Riemann, the German mathematician who in 1859 introduced what became known as the Riemann zeta function. He suspected the function was deeply tied to how prime numbers, those divisible only by one and themselves, are spread across the number line.
If Riemann's intuition was correct, the apparently random scattering of primes hides an underlying order. Mathematicians have tried to establish that for the 167 years since, without success until this week's claim.
Why the Hypothesis Matters So Much
The problem's importance goes well beyond prime numbers. It links many branches of mathematics that otherwise appear unrelated, acting as a kind of connective tissue for the discipline.
A large body of existing mathematical work depends on it. Numerous proofs and conjectures are valid only on the assumption that the Riemann hypothesis is true, so a confirmed proof would firm up a great deal of research at once.
An Unusual Announcement Backed by Formal Code
Alpöge shared the proof on his personal website in a file titled "remark.pdf", an understated way to present a potential landmark result. His accompanying comments were far more effusive, crediting Claude with racing ahead of him in the final stretch of the work.
The enthusiasm came with substantial supporting material: roughly 17 million lines of code written in Lean, a programming language designed to verify that each step of a proof is logically sound. Formal verification of this kind could make it easier for others to test whether the argument holds.
Alpöge also says the work goes further than the original problem. He claims he and Claude have proved the generalised Riemann hypothesis, which extends the question to a broader family of mathematical functions.
Part of a Wave of AI-Assisted Breakthroughs
The result does not stand alone. Alpöge has been involved in several of the high-profile AI-enabled mathematical advances of recent months, which lends some credibility to his claim even before formal review.
Since September, hundreds of long-standing open problems have reportedly been resolved with AI help, leaving researchers with thousands of pages of machine-generated proofs to examine. On October 6th, OpenAI announced that one of its models had solved the quasi-Riemann hypothesis, seen as an important step toward the full problem.
Jonathan Keating, also of the University of Oxford, called the new result a very big leap beyond that. He said the quasi-Riemann proof had already stunned him, but described this latest claim as several orders of magnitude more surprising.
Verification Comes Next
It remains too early to declare the problem solved. Mathematicians are expected to spend the coming days scrutinising the proof, and peer review of a result of this scale typically takes considerable time.
Beyond whether the proof is correct, many researchers will want to understand how the AI arrived at it. The method could prove as significant as the result itself for the future of mathematical research.
What It Means for AI and the People Building It
For Anthropic, the timing is notable. The claim lands just after its record IPO, at a moment when financial markets have been unsettled, and offers a striking demonstration of what its models may be capable of.
For the wider industry, the episode underlines a shift in how frontier AI labs are competing: increasingly on their ability to contribute to hard scientific problems rather than only on consumer products. If the proof holds, it would mark one of the clearest cases yet of AI acting as a genuine research collaborator, and it will raise new questions about how such work is checked, credited and understood.
Source: Livemint





