
A mathematician working at Anthropic says he has solved the Riemann hypothesis, the century-and-a-half-old problem considered math's most famous unsolved puzzle, with help from the company's own AI system. Levent Alpöge posted a four-page note on his personal website declaring the breakthrough, writing, “I believe I have, working with Claude, discovered a proof of the Riemann Hypothesis.” Within hours, a second and entirely separate proof attempt also surfaced, setting off a scramble among mathematicians to figure out which claim, if either, actually holds up.
What the Riemann Hypothesis Actually Says
The Riemann hypothesis, formulated in 1859 by German mathematician Bernhard Riemann, concerns the distribution of prime numbers along the number line toward infinity, according to The New York Times. Riemann hypothesized a formula for estimating how many primes exist below any given number, a pattern that becomes harder to pin down as numbers grow larger — primes make up roughly 25 percent of whole numbers up to 100 but only about 7.8 percent of whole numbers up to one million, per the same report. Formally, the hypothesis asserts that all non-obvious zeros of the zeta function are complex numbers with real part one-half, according to the Clay Mathematics Institute. It is one of the institute's seven Millennium Prize Problems, with $1 million allocated to the solution of each, and it also appeared among the 23 unsolved problems that German mathematician David Hilbert laid out in 1900.
Alpöge linked computer code written in the proof-assistant language Lean, which he said would verify the result. Lean turns mathematical statements into a form that can be checked for logical consistency, and clean compilation is generally taken as a sign the underlying logic holds, the Times reported. Other researchers immediately began examining that code, and MIT mathematician Andrew Sutherland was among those who independently compiled Alpöge's result in Lean without issues, according to Scientific American.
Experts Split Between Confidence and Caution
Sutherland said the Lean kernel independently rechecked the proof using three standard axioms, leaving him little room for doubt. “Honestly the chance that anything is fishy here is approximately zero,” he said, per that outlet's report. Separately, a clean build using standard axioms from the Lean 4 Dev project confirms that Lean's kernel accepts a proof of the stated theorem — though, as the project notes, “it does not yet tell anyone why the argument works.”
Not everyone is ready to declare victory. Researchers are still checking whether the proof relies on any hack, dodge, or shortcut that would quietly change the meaning of the hypothesis or bend other mathematical rules, the same Scientific American report noted. Mathematician Kevin Buzzard said a crucial step remains outstanding: “We still await a successful comparator run” to confirm the code's behavior and compilation across independent kernels, he said.
Cornell mathematician Steven Strogatz described the Riemann hypothesis as nothing less than “the Mount Olympus” of the field, and said Alpöge played a central role in guiding Claude toward the mathematical ideas behind the proof, per the Times. Rutgers mathematician Alex Kontorovich said he had expected not to be rattled by high-level AI mathematics at this point — and was shocked anyway. James Maynard went further, saying he thinks Alpöge's proof is likely correct and putting his own confidence at 98 or 99 percent, while cautioning that the Lean verification still needs to be checked carefully to make sure it proves the intended claim rather than some adjacent statement. “People want to understand why the Riemann hypothesis is true,” Maynard said, adding that a confirmed proof would deliver that understanding and be “worth celebrating.”
A Rival Claim Complicates the Picture
Hours after Alpöge's announcement, a second and unrelated proof claim emerged, according to a separate Scientific American report. Alessandro Masullo, an independent mathematician living in Rome who has worked on the problem for years, released an incomplete proof of the Riemann hypothesis built without any artificial intelligence, according to Brian Conrad, the Stanford mathematician who was Masullo's doctoral thesis adviser. Conrad said Masullo's proof uses a different approach than Alpöge and Claude's and called it credible, without major gaps — but Conrad also said Alpöge released a proof he does not fully understand, and he flagged broader unease about the security of information entered into AI systems like Claude, per the Times.
Masullo's draft is, by multiple accounts, unfinished. Mathematician Ravi Vakil reported that the paper is “undergoing major revision” and that nobody has verified it, according to Nacionale. Conrad has been reviewing the work for months and has given extensive feedback that is being incorporated, according to the Math 216: Foundations of Algebraic Geometry blog; Conrad has also said Masullo is revising the writing and fixing minor errors, and plans to release a polished paper sometime in the coming year. Alpöge did not reply to the Times' request for comment on his own proof.
Part of a Bigger AI-Math Moment
The dueling claims land amid what Nature described as a year in which artificial intelligence has reached one spectacular mathematical breakthrough after another. In September, OpenAI announced it had solved a longstanding Navier-Stokes question — the equations that describe the flow of liquids — with a proof that found a case in which a specially forced liquid could accelerate to infinite speeds, according to the Times. Alpöge himself had already been active in that space, revealing a proof of a different variation of the Navier-Stokes problem alongside New York University mathematics professor Tristan Buckmaster; the pair claim a liquid could reach infinite speeds even without a driving force.
Then in October, OpenAI released more than 350 findings, among them a proof of the so-called quasi-Riemann hypothesis — a version of the problem with looser constraints that is easier to prove, announced on October 6, according to Quanta Magazine. Mathematicians studying that result largely agreed it was correct, though they described its novelty as lying in how it stitched together existing number-theory ideas rather than breaking genuinely new ground. Whether Alpöge's full-strength Riemann claim, or Masullo's human-built alternative, ultimately clears that same bar of scrutiny remains, for now, an open question that the broader mathematics community is still working through.









