Technology

Anthropic mathematician claims Riemann hypothesis proof as independent code checks advance

Levent Alpöge, a mathematician at Anthropic, says he and the company's Claude model have proved the Riemann hypothesis. Independent mathematicians reported on November 20 that they had compiled and rechecked the Lean formalization, but another technical check remained pending, and researchers cautioned that understanding the argument would take much longer.

lake-uankxtpl20t15euqi6743503-3a646cb7-8fe7-402e-8ec9-3b9604339873-0.webp

Early checks and a long review ahead

Alpöge announced the claim in a four-page document containing a proof sketch. "I believe I have, working with Claude, discovered a proof of the Riemann Hypothesis," he wrote. The claim concerns what many mathematicians regard as their field's most important unsolved problem.

Scientific American reported that independent mathematicians had separately compiled and checked the formalization on November 20, 2026. MIT mathematician Andrew Sutherland said it "has been compiled and independently re-checked by the Lean kernel with only the three standard axioms" and that "the chance that anything is fishy here is approximately zero."

A further check was still awaited, according to Kevin Buzzard of Imperial College London. Known as a "comparator run," it checks whether the code redefines symbols or exploits a kernel bug. Buzzard said this typically takes a matter of hours.

James Maynard, a number theorist and Fields Medal winner at the University of Oxford, expressed optimism while keeping his assessment conditional. "If this fully checks out, it is clearly a historic milestone in mathematics," he told Nature. He added: "I suspect that the proof will turn out to be correct."

What the claim would establish

Bernhard Riemann formulated the hypothesis in the mid-1800s while studying how prime numbers are distributed. The German mathematician's birth bicentenary fell in September 2026.

The hypothesis concerns the zeros of the Riemann zeta function: points z in a plane where ζ(z) = 0. Those zeros encode information about the distribution of primes. Riemann proposed that, aside from certain straightforward exceptions, they all sit on a particular vertical line. He showed that this would mean primes are distributed among whole numbers in an essentially maximally random way.

Alpöge's claim goes further than the original hypothesis. He says the method also proves the Generalized Riemann Hypothesis for Dirichlet L-functions. Resolving the Extended Riemann Hypothesis remains a prospect: he said he expects the approach could do that with more time.

In his account of the method, Alpöge pointed to Sergei Stepanov's 1969 proof of a similar statement in a simpler branch of mathematics. He said the new argument uses weighted prime sums, called von Mangoldt sums, along with constructions known as theta lifts.

A vast formal proof, not a conventional paper

Beyond the short sketch, the proof was released as an archive of Lean code rather than a conventional mathematical paper. Interesting Engineering and Hindustan Times reported that the archive contains approximately 175,187 source files and 239 megabytes of data. Alpöge said a machine-generated natural-language draft exceeded 250,000 pages.

Alpöge acknowledged that he is not yet an expert in the portions Claude developed. In a footnote, he warned that his explanation, based on a machine-written summary, "might make zero sense"; he nevertheless expected the argument to be "massively simplified, likely immediately."

Researchers distinguished checking the formalization from understanding the mathematics. Maynard described the current Lean code as incomprehensible. Tristan Buckmaster of New York University told Nature: "At the moment, it apparently is crazily complicated. It's going to take a long time for mathematicians to digest."

Maynard agreed that understanding would take time, adding: "assuming we do get the corresponding understanding, this will surely be viewed as one of the absolute pinnacles of mathematical results."

Criticism of the release and its timing

Andrew Granville, a number theorist at the University of Montreal, criticized the decision to announce the result before preparing a clearer explanation. "What he released is garbled," he told Nature.

The announcement followed Anthropic's initial public offering by just days. The San Francisco company's IPO initially put its valuation at $1.5 trillion, but that valuation fell on November 19. Granville attributed the timing of the proof's release to the stock decline.

Alpöge rejected that interpretation in his document: "There is an obvious cynical view, to do with the timing, but it is false." The Verge reported that he had texted Anthropic CEO Dario Amodei about the solution "late Sunday night" and suggested that he delayed publication until after the IPO. Anthropic did not immediately answer Nature's request for comment.

The claim also contrasts with Anthropic's earlier expectations. In August 2026, the company disclosed work on the problem with Claude but said: "We don't expect that the techniques Claude used will lead to proving the Riemann hypothesis."

Earlier AI work and the path to recognition

Buckmaster, who worked with Alpöge on a separate Navier-Stokes project, credited him with developing techniques for obtaining mathematical results with large language models. "Levent himself built this amazing setup to do math," he said. Alpöge is a number theorist on leave from Harvard University; according to Buckmaster, he had spent months working intensely on this and other problems, with very little sleep.

The Verge reported that Alpöge had already used Claude for other mathematical results in 2026, including a claimed solution to the Jacobian conjecture. Alpöge wrote that solving the Riemann hypothesis had been an ambition since his student days and credited Claude with producing "an astounding amount of mathematics." He described the experience as both the loss of a guiding goal and the possible beginning of something much larger across science.

Another major AI-assisted claim came from OpenAI in September 2026, when it said it had solved the Navier-Stokes problem. According to The Verge, that effort cost millions of dollars and used 10,000 agents. The result is widely considered credible but has not received formal recognition. It has also faced allegations of questionable research practices, including claims that OpenAI used mathematicians' unpublished work.

Both Navier-Stokes and the Riemann hypothesis belong to the Clay Mathematics Institute's seven Millennium Prize Problems, designated in 2000 with a $1 million award for each. Only the Poincaré conjecture has been officially solved.

The institute's rules require a proposed solution to appear in a qualifying outlet, undergo at least two years of scrutiny and win general acceptance from the mathematics community before it can be considered for a prize.

IPOs

Share

Twitter/XFacebookLinkedIn

Read also