A mathematician from Anthropic says he has found a proof of the Riemann hypothesis, one of the most famous and intractable problems in mathematics. It was first formulated by the German mathematician Bernhard Riemann in 1859 and assumes that the distribution of prime numbers follows a certain pattern. This problem is included in the list of “Millennium Problems” of the Clay Mathematics Institute, for its solution there is a reward of $1 million.

Image source: unsplash.com
Levent Alpöge from Anthropic reported that he found a solution to the Riemann hypothesis using the Claude neural network. It is written in Lean, a tool that mathematicians use to check proofs by machine. This process, called formalization, is a sequential test of logical steps. It is so thorough that if the theorem passes the test (that is, the code compiles successfully), this is considered proof of its correctness beyond reasonable doubt.
This method has its own vulnerabilities, so both the code itself and the mathematical calculations must be checked in detail by people. “First, let’s see if the code compiles,” Kevin Buzzard from Imperial College London commented on the news. -It may take time to find experts who can explain the essence of the evidence – provided, of course, that it is correct. And it is most likely true: after all, if the code in the Lean system does not compile or turns out to be imperfect in some way, Alpöge will find himself in an extremely awkward position.”
Alpöge deliberately avoided publishing a human-readable version of the proof, noting that it would be “completely inappropriate to publish a bulky, unreadable and meaningless set of data.” However, he mentioned that the unpublished version of the proof totals more than 250,000 pages. In August, Anthropic already reported on developments regarding the Riemann hypothesis, but then it was still far from a full proof.
“I believe that, working together with Claude, I have found a proof of the Riemann hypothesis,” Alpaege said. “It’s natural to feel that this is the end—in my case, the loss of the main guiding light—but what I saw in the process of work amazed me so much that, it seems to me, this is the beginning of something much more significant for all of science.”

Alpaege notes that his proof could help solve the generalized Riemann hypothesis (GHR)—a statement that extends the concept of the Riemann hypothesis to other properties of prime numbers—and also contribute to a better understanding of a broader class of mathematical structures.
“There are many results that until now were only conditional – depending on the generalized Riemann hypothesis – and now they have become full-fledged theorems! – says James McKee from Royal Holloway University of London. “What is truly amazing—and at the same time alarming—is the rapid progress in recent months in the field of automated proof of significant mathematical results. It’s an amazing, but at the same time frightening time for a research mathematician.”
The essence of the Riemann hypothesis lies in a special zeta function. It can be thought of as an infinite sum or as a tool for studying the distribution of prime numbers that are divisible only by 1 and themselves. Riemann suggested that if you take all the non-trivial zeros of the zeta function, they will be located on the same vertical line in the complex plane – with the real part equal to 1/2.

The Riemann hypothesis asserts the orderliness of chaos. The prime numbers appear to be scattered randomly, but Riemann suggested that there was a logic to their distribution. We are talking about the very nature of numbers – this is a fundamental question, similar to the question about the structure of the universe in physics.
The Riemann hypothesis underlies many calculations. All modern cryptography depends on understanding the behavior of prime numbers. Almost all modern digital locks (bank cards, passwords, secure chats) use operations with large prime numbers. Many computational methods and algorithms depend on the speed and accuracy of their finding. Hypotheses have also been put forward about the possible connection of the statistical properties of the nontrivial zeros of the Riemann zeta function (and therefore prime numbers) with the phenomena of quantum physics, in particular with quantum chaos.
Previously, it was confirmed by numerical methods that more than ten trillion of the first nontrivial zeros of the Riemann zeta function satisfy this hypothesis, which is a good argument in favor of the truth of this hypothesis, but does not guarantee it. Even one deviation will be enough for the entire hypothesis to collapse.
Alpöge has already been involved in a number of recent AI-related discoveries. In July, he announced that the Jacobian hypothesis, which scientists had struggled to prove for decades, was in fact false, citing a counterexample just 216 characters long to refute it. In addition, he and Tristan Buckmaster of New York University solved three important problems that were considered “stepping stones” toward solving the Navier-Stokes problem. This came just hours before OpenAI unveiled a complete solution to the problem.
Anthropic went public on November 19, and Alpäge appears to have deliberately delayed announcing his proof until then. “The timing is extremely unfortunate. If I could, I would have finished everything much earlier,” he noted. According to him, in the process of discussing the situation with the main Anthropic Dario Amodei, it was decided to “observe certain restrictions” before publication.
If you notice an error, select it with the mouse and press CTRL+ENTER.


