Anthropic Researcher Says AI Has Proved the Riemann Hypothesis, Maths' Greatest Unsolved Problem
A mathematician at Anthropic says he has used the company's AI system Claude to prove the Riemann hypothesis, the most famous unsolved problem in mathematics, in a claim that has not yet been independently verified.
Levent Alpöge, a mathematician employed by Anthropic, announced on Friday that he believes he and Claude, the company's artificial intelligence system, have discovered a proof of the Riemann hypothesis. He also said the same approach settles the generalised Riemann hypothesis, a broader version of the problem. Instead of publishing a conventional paper, Alpöge released the argument as a formal proof in Lean, a programming language that lets computers check each step of a mathematical argument.
The claim has not been independently verified, and mathematicians around the world are now working through the material. James Maynard, an Oxford professor and winner of the Fields Medal, said that if the proof holds up it would be a historic milestone, adding that he suspected it would turn out to be correct, although checking the computer verification alone would take some time. A mathematician at MIT independently compiled the released code shortly after it appeared.
The announcement came a day after Anthropic staged the largest stock market debut in history, raising $81.3bn and reaching a valuation of about $2 trillion. Alpöge said he had spoken to Anthropic chief executive Dario Amodei before making the announcement, and the company declined to comment. Hours later, a second claim appeared: a draft by Alessandro Masullo arguing for the generalised hypothesis by a wholly human route, shared by the president of the American Mathematical Society, Ravi Vakil, after months of private review.
Bernhard Riemann first set out the hypothesis in 1859. It concerns how prime numbers, the building blocks of arithmetic, are distributed. It is one of seven Millennium Prize Problems set by the Clay Mathematics Institute, each carrying a $1m prize, and it has resisted every attempt at a proof for 167 years. In August, Anthropic said an early Claude model had advanced work on a related Riemann result, while making clear the hypothesis itself remained unproved, and in September an OpenAI model was reported to have cracked a different Millennium problem, a result that also awaits full vetting.
Key points
- Anthropic mathematician Levent Alpöge says he and the AI system Claude have proved the Riemann hypothesis, and the generalised version too, releasing the argument as computer-checkable Lean code.
- The claim has not been independently verified; experts are checking the work, with one Fields Medal winner calling it a possible historic milestone if it holds up.
- A second, separate claim to the generalised hypothesis by a human mathematician surfaced the same day, and is also awaiting verification.
- The hypothesis, proposed in 1859, concerns the distribution of prime numbers and is one of seven $1m Millennium Prize Problems; the announcement came a day after Anthropic's record stock market debut.