In a four-page note published on his website, Alpöge described how he used Claude to work through the mathematical problem. The effort reportedly involved generating a detailed argument and formal proof files that can be examined using Lean, a programming language designed to verify mathematical proofs.
What is the Riemann hypothesis, and why does it matter?
The Riemann hypothesis is a conjecture about the distribution of prime numbers – numbers divisible only by themselves and one.
Prime numbers such as 2, 3, 5, 7, 11 and 13 form the building blocks of number theory. Although mathematicians have discovered many properties of primes, their distribution among increasingly large numbers follows patterns that remain difficult to explain completely.
Riemann proposed that these patterns could be understood through a mathematical function known as the Riemann zeta function.
The hypothesis states that every non-trivial zero of this function has a real part equal to one-half.
The significance extends well beyond an abstract mathematical puzzle. The hypothesis is closely connected to how accurately mathematicians can estimate the distribution of prime numbers. Its resolution could have implications for number theory, mathematical analysis and areas of computer science that rely on the properties of primes.
It is also one of seven Millennium Prize Problems identified by the Clay Mathematics Institute in 2000. A correct solution carries a prize of $1 million.
How did Claude help develop the claimed proof?
Alpöge's announcement raises a significant question about the changing relationship between mathematicians and AI systems.
Rather than presenting Claude simply as a tool for answering mathematical questions, the effort involved using the chatbot as a collaborator over several weeks to develop and examine a complex mathematical argument.
The process reportedly took around six weeks, with Alpöge guiding the work and using Claude to advance the proof. The released material includes a mathematical note and formal proof files intended for verification.
One important aspect is the use of Lean, a proof assistant that allows mathematical arguments to be translated into a formal language that computers can check.
This matters because AI-generated mathematical reasoning can sound convincing while containing logical gaps or unsupported assumptions. Formal verification offers a way to test whether each step follows from the preceding ones.
Nevertheless, a computer checking a formal proof does not automatically establish that the underlying mathematical argument is correct. Researchers must also examine the formalisation, its assumptions and whether it genuinely establishes the original conjecture.
What does Lean verification mean?
Lean is an open-source proof assistant used by mathematicians to check complex mathematical arguments. It allows researchers to express mathematical statements in a formal language and verify whether the logical steps establish a specified conclusion.
Alpöge has released a large collection of Lean files alongside his four-page note. The material reportedly contains more than 175,000 source files, with the formalisation running to an enormous length.
This is an important distinction. If the code compiles, it indicates that the formal proof system has checked the encoded argument. But mathematicians must still establish that the theorem encoded in the files is genuinely equivalent to the Riemann hypothesis and that no errors or hidden assumptions undermine the result.
In other words, passing a computer-based verification check is an important milestone, not the final scientific verdict.
Why the timing of the announcement is significant
The announcement arrived on November 20, just a day after Anthropic began trading on the US stock market.
The timing has drawn attention because the company is competing with OpenAI, Google and other technology firms to demonstrate increasingly advanced AI capabilities.
Alpöge, however, presented the work as a mathematical development rather than a corporate product announcement. He said the process had taken place over approximately six weeks and described Claude as having made particularly striking contributions towards the end.
He also claimed that the approach could extend beyond the original Riemann hypothesis to the Generalized Riemann Hypothesis, a broader conjecture concerning the zeros of Dirichlet L-functions.
If validated, that would make the claim even more consequential for number theory.
Is AI beginning to solve problems that mathematicians could not?
The announcement comes amid growing interest in the use of AI systems for advanced mathematical research. Traditional AI chatbots are primarily designed to generate responses, explain concepts and assist with coding.
Mathematical research, however, requires a different level of reasoning: developing original conjectures, identifying connections between mathematical ideas and producing arguments that withstand rigorous scrutiny.
Alpöge's account suggests that Claude was not merely helping with calculations or explaining established mathematical concepts. He credited the chatbot with developing crucial parts of the argument and said it had outpaced him towards the end.
That distinction could be important for the future of AI-assisted scientific research.
AI systems are increasingly being tested as research collaborators capable of exploring mathematical approaches that may take human researchers years to develop. However, their ability to generate sophisticated arguments does not eliminate the need for human judgement, independent verification and reproducibility.
Another mathematician enters the race
The developments have become more interesting with another mathematician announcing a separate attempt to solve the same problem.
Alessandro Masullo, a postdoctoral researcher at Stanford University, released a proof attempt that he said he had developed independently over 12 years, without using AI.
Masullo said he had been approximately a year away from publishing his work but decided to release the current version following Alpöge's announcement. He cautioned that his draft could contain errors because it had been rushed.
The emergence of two competing claims has intensified the debate over whether AI is beginning to transform mathematical research and what role human mathematicians will play in that process. Neither proof had been formally established as correct as of November 20.
What happens before the Riemann hypothesis can be declared solved?
The biggest hurdle for Alpöge is independent mathematical verification. Several mathematicians have reportedly confirmed that the submitted Lean code compiles. However, this does not settle whether the argument correctly proves the Riemann hypothesis.
Mathematicians must examine the underlying logic, check the formalised statement against the accepted mathematical definition and establish that the proof does not depend on unsupported assumptions.
There is also the question of producing a comprehensible mathematical paper. Alpöge has acknowledged that the argument is currently extremely long and said he intends to understand it more fully, simplify it and present it in a form that other mathematicians can examine.
The Clay Mathematics Institute's rules also require a proposed solution to be published in an eligible peer-reviewed publication and to withstand at least two years of scrutiny before it can qualify for the Millennium Prize. For now, the Riemann hypothesis remains an open mathematical problem.