Levent Alpöge, a Turkish American mathematician at Anthropic, says the company’s AI software, Claude, helped him prove the Riemann hypothesis. Other mathematicians have begun checking his work, but the result has yet to be confirmed.

In a note on his personal website on Friday, Alpöge said the work took about six weeks. He credited Claude with outpacing him in the final stages. He also linked to code in the programming language Lean that he said could verify the result.

Anthropic had gone public on Nasdaq the day before the note appeared. Alpöge called the timing “horrible” and said he wished he could have finished sooner.

German mathematician Bernhard Riemann proposed the hypothesis in 1859. It concerns the distribution of prime numbers: whole numbers greater than one whose only positive divisors are one and themselves. Riemann developed a formula to estimate how many primes lie below a given number. The Clay Mathematics Institute has offered $1 million for a solution to each of its seven Millennium Prize Problems, including the Riemann hypothesis.

If Alpöge’s proof holds up, it could mark a milestone for AI in mathematical research. Some mathematicians remain wary of relying on AI for such work. On Friday, Alessandro Masullo, an independent mathematician in Italy, released an incomplete proof of the Riemann hypothesis, according to the New York Times. Masullo said he had worked on it for nearly 12 years and had not used AI because he did not fully trust it.

The cost of these efforts has also drawn criticism. In September, OpenAI said an unreleased version of its AI tools had helped solve part of a different Millennium Prize Problem. Alpöge was working on that problem for Anthropic at the time. OpenAI spent $15 million to put 10,000 AI systems to work on it.

James Robinson, a mathematician at the University of Warwick, told the Guardian in September that he objected to tech companies burning fuel to show off their latest models. He warned that such competition could cause substantial environmental harm.