Anthropic mathematician claims Claude AI helped prove the Riemann hypothesis
Levent Alpöge, a mathematician at Anthropic, says he used a non-public version of the Claude chatbot to prove the Riemann hypothesis. Experts have not yet verified the claim, and checking it will take time.
This short update was prepared automatically from other news outlets’ reporting and has not yet been reviewed by an editor.
An Anthropic employee who is also a mathematician says he has used the company's AI chatbot to prove the Riemann hypothesis, one of the best-known unsolved problems in mathematics. The claim has not been verified.
According to BBC News, Levent Alpöge announced the result on Friday, one day after Anthropic's debut on the US stock market. He said he spent about six weeks working with a non-public version of Claude to build a detailed proof. If it is correct, the BBC notes, it would rank among the most significant proofs ever produced.
The hypothesis is almost 170 years old and is named after the mathematician Bernhard Riemann. It concerns prime numbers and proposes that they follow a fixed order along the number line. Primes underpin higher mathematics, so progress on the problem could have wide effects.
The proof is written in Lean, a computer language used to check mathematical proofs, and runs to more than 250,000 lines.
Alpöge said he messaged Anthropic chief executive Dario Amodei on Sunday to ask how he could make the discovery public. He admitted the timing alongside the stock listing was unfortunate and said he would have finished earlier if he could. He said he alone was responsible for deciding he could not wait any longer. He acknowledged that some people might take a cynical view of the timing, but called that view false.
Prominent mathematician Terrance Tao said on Friday that verification is still a long way off. He expected the process of working through the proof to take time, but noted that in this case there is a human author to help with the task.
The BBC says major AI-assisted maths results earlier this year had no such human involved. In September, OpenAI said staff using a non-public version of its tools had solved part of a notoriously hard problem on the Navier-Stokes equations, which describe how fluids move, in 88 hours. More than 10,000 AI agents were assigned to the task as a training exercise. That proof has still not been confirmed, and it drew criticism from academic and mathematical communities.
A group of mathematicians wrote in a joint letter that AI firms advertising big solutions to show off their commercial products harms the science of mathematics. A month later, OpenAI released what it described as solutions or significant progress on 372 more open problems. The BBC says expert review of those claims could take months or longer. Anthropic has also claimed progress in mathematics and physics.

Comments (0)