Claude AI Helps Prove Riemann Hypothesis, Math's Top Prize

Published (Updated )

The quick answer
- Levent Alpöge, a mathematician at Anthropic, announced on Friday, November 20, 2026, that he believes he has proved the Riemann hypothesis, working with the company's Claude AI model.
- The Riemann hypothesis, first posed in 1859, is the most famous unsolved problem in mathematics, and it is one of the seven Millennium Prize Problems, each carrying a $1 million prize.
- Alpöge published his proof in Lean, a programming language that checks mathematical arguments step by step, and the proof compiles without errors.
- Independent mathematicians have now separately compiled the proof in Lean, including Andrew Sutherland at the Massachusetts Institute of Technology.
- Sutherland said the Lean proof had been independently rechecked by the Lean kernel using only the three standard axioms, and that the chance anything was wrong with the formal check was approximately zero.
- A final verification step, called a comparator run, is still pending, and mathematicians say it could take months or even years to fully understand the proof.
- A separate, fully human proof of the generalised Riemann hypothesis was also posted this week by a Stanford postdoctoral researcher, Alessandro Masullo.
- The claim is not yet a confirmed result, but the early independent checks make it the most credible claimed solution in the problem's 167-year history.
What was actually announced
- Alpöge posted a statement on his personal website on Friday morning, November 20, 2026, saying he believed he had discovered a proof of the Riemann hypothesis, working with Claude.
- He described the moment as the end of a childhood obsession, and said he had messaged Anthropic chief executive Dario Amodei late on Sunday night to ask how to announce the result.
- The proof was released not as a conventional mathematical paper but as code in Lean, a proof assistant that verifies each step of an argument mechanically.
- Alpöge's statement also says the proof covers the generalised Riemann hypothesis, a much larger claim that extends the original idea to a whole family of mathematical functions.
- He released a short four-page document sketching the argument, alongside the Lean code itself.
- The timing was awkward, and Alpöge acknowledged it. Anthropic had staged the largest initial public offering in history the day before, on Thursday, November 19, and the company's shares then fell sharply on Friday.
- Alpöge wrote that there was an obvious cynical view about the timing, but that it was false. The work had come together late on Sunday night, before the IPO, and he said he could not live with himself if he waited to release it.
- Anthropic did not immediately respond to requests for comment from the press.
- The announcement landed in a year of extraordinary AI progress in mathematics. An internal OpenAI model reportedly solved another Millennium Prize Problem in September 2026, and in 26 years humans alone had solved only one of the seven.
What the Riemann hypothesis actually says
- The Riemann hypothesis concerns the distribution of prime numbers, the numbers divisible only by themselves and 1, which are the building blocks of arithmetic.
- It was proposed in 1859 by the German mathematician Bernhard Riemann, who was born 200 years ago this September.
- The hypothesis is about a function called the Riemann zeta function, which encodes the pattern, or apparent randomness, of the primes.
- In simple terms, it says that all the interesting zeros of that function lie on a single straight line, known as the critical line.
- If that is true, the primes follow a hidden order. If it is false, much of modern number theory would need to be rethought.
- Thousands of mathematical results are proved on the assumption that the hypothesis is true, which is why mathematicians call it the most important open problem in the field.
- The generalised version extends the same claim to a broader family of functions, and it underpins even more of mathematics, which is why Alpöge's claim to have proved the generalised version is so significant.
- The Clay Mathematics Institute offers $1 million for a proof of the Riemann hypothesis, as one of its seven Millennium Prize Problems, set in the year 2000.
- No human has proved it in 167 years, and no computer has proved it either, until now, if the new claim holds.

How the proof was checked, and what is still missing
- The first check was mechanical. Lean is a programming language and proof assistant that verifies the logic of each step in a mathematical argument.
- If a Lean-coded proof compiles without errors, that is usually taken as confirmation that the logic and mathematics in each step are correct.
- Alpöge's proof compiles, and that alone is a strong signal, but it is not the whole story.
- The first question mathematicians ask about a Lean proof is whether the statement of the theorem in the code actually corresponds to the real theorem.
- Kevin Buzzard, a mathematician at Imperial College London, explained in a blog post that this is the first thing to check when judging a document that comes with Lean code.
- The second question is whether the code compiles for other mathematicians, on their own machines, with their own copies of Lean.
- That question now has an answer. Independent researchers, including Andrew Sutherland at MIT, have separately compiled Alpöge's result in Lean without issues.
- Sutherland said the Lean proof had been compiled and independently rechecked by the Lean kernel with only the three standard axioms, and that the chance anything was fishy was approximately zero.
- A third check is still pending. Researchers are running a comparator run, which checks that the code does not do anything nefarious, such as redefining the meaning of any of the symbols in the statement.
- The comparator run also checks, in paranoid mode, that the code compiles with several independent kernels, so it is not exploiting an unknown bug in Lean's official kernel.
- That process typically takes a matter of hours, and researchers said on Friday that they were still awaiting a successful comparator run.
- Even after the formal checks pass, a deeper question remains. Mathematicians will need to understand how the proof works before broadly accepting it, and that could take months or even years.
- Youness Lamzouri, a professor of mathematics at the University of Lorraine in France, cautioned that checking the computer-generated proof was only the beginning.
- James Maynard, the Regius Professor of Mathematics at the University of Oxford and a 2022 Fields Medal winner, said he suspected the proof would turn out to be correct, even though it was currently in the form of incomprehensible Lean code.
- Maynard told reporters that if the result fully checked out, it was clearly a historic milestone in mathematics, establishing one of the great mysteries of the subject.
- Andrew Granville, a number theorist at the University of Montreal, was more sceptical, saying the four-page sketch was garbled and that Alpöge should have prepared a clearer write-up before announcing.
- Granville also noted the timing, pointing out that Anthropic's valuation had dropped on November 19, though Alpöge has denied any connection.
The human proof that appeared at the same time
- While Alpöge's AI-assisted proof was making headlines, a second proof appeared from a different direction.
- Ravi Vakil, a professor at Stanford and president of the American Mathematical Society, posted a proof of the generalised Riemann hypothesis written by Alessandro Masullo, a Stanford postdoctoral researcher.
- Vakil noted that Masullo's proof was all human, with no AI involved.
- The two claims are different in character. Alpöge's proof is a formal Lean argument that a machine can check but humans cannot yet read easily. Masullo's is a human-written mathematical argument.
- It is not yet clear whether the two proofs are related, or whether they use similar ideas, and mathematicians are still reading both.
- The coincidence is remarkable. A problem that resisted every mathematician for 167 years may now have two claimed proofs in the same week, one from a human working with AI and one from a human working alone.
- For the mathematics community, the practical effect is the same either way. Both claims need careful checking, and the checking will take time.
- Alpöge has committed to writing up his proof properly, in a form humans can understand, when he is able to simplify it.
- He described the experience of finding the pieces of the argument with Claude in euphoric terms, saying the model flew and danced and sparkled near the finish line.
- That language has divided opinion. Some mathematicians see it as a genuine account of a new kind of collaboration. Others see it as hype around a result that has not yet been digested.
Why AI is suddenly solving maths problems
- The Riemann claim did not come from nowhere. It is the latest in a run of AI-assisted breakthroughs in mathematics in 2026.
- In September, an internal OpenAI model reportedly solved a Millennium Prize Problem, the first of the seven to fall to AI, after using millions of dollars of computing power.
- In October, one of the proofs in a large dump of mathematical results from OpenAI made significant progress on understanding the primes, the same territory as the Riemann hypothesis.
- In the summer, a race began between frontier AI labs and mathematicians to solve outstanding problems, and Alpöge and his collaborator, Tristan Buckmaster of New York University, also released an important proof related to the Navier-Stokes equations, another Millennium Prize Problem.
- The pattern is consistent. AI models are being used to search for arguments, test ideas and fill in gaps at a speed no human team can match, and the results are then formalised in Lean so they can be checked.
- In 26 years, humans alone had solved one Millennium Prize Problem. In three months, AI now appears to have solved two, if the claims hold.
- The shift has consequences beyond mathematics. If AI can solve the hardest open problems in a field, it raises questions about what human expertise is for, and how science will be done in the future.
- It also raises questions about verification. A proof that only a machine can read is a strange object, and mathematics has always been a human activity, built on understanding.
- The Riemann hypothesis is a test case for that tension. The Lean code may be correct, and independent checks suggest it is, but until humans understand why it is correct, the result will feel unfinished.
- That is why the mathematicians quoted in the coverage are careful with their words. They say the proof is almost certainly true, and that it is a historic milestone if it checks out, but they also say the work of understanding it has barely begun.

What happens next
- The immediate next step is the comparator run, which should confirm that the Lean code states the real Riemann hypothesis and does not exploit a bug in the proof checker.
- After that, the Clay Mathematics Institute's rules come into play. A Millennium Prize is only awarded after a proof has been published in a refereed journal and has survived two years of scrutiny by the mathematical community.
- That means no prize money is coming soon, whatever the comparator run says. The two-year clock has not even started.
- Alpöge has said he plans to write up the proof in a human-readable form, and that process will be watched closely, because it is the only way the wider community can judge the ideas behind the code.
- Masullo's human proof will go through the same process, and mathematicians will compare the two approaches to see whether they share a common idea.
- The wider question is whether AI-assisted proof becomes normal in mathematics. If the Riemann claim holds, it will be hard to argue that AI is only a tool for routine calculation.
- For now, the honest summary is the one the experts are giving. The proof compiles, independent mathematicians have rechecked it, and the chance of a formal error is close to zero. But full understanding, and the prize, are still a long way off.
- The Riemann hypothesis has survived 167 years of attacks by the best mathematicians in history. It may finally have fallen, and it may have fallen to a mathematician working with an AI. That, by itself, is a story worth telling carefully.
The bottom line
- Anthropic mathematician Levent Alpöge says he and Claude have proved the 167-year-old Riemann hypothesis, and the generalised version too.
- The proof is written in Lean, it compiles, and MIT's Andrew Sutherland has independently rechecked it, saying the chance of anything fishy is approximately zero.
- A comparator run is still pending, and mathematicians say it could take months or years to fully understand the proof.
- A separate human proof of the generalised hypothesis was posted by Stanford postdoc Alessandro Masullo.
- The Riemann hypothesis is a $1 million Millennium Prize Problem, but the prize requires publication and two years of scrutiny, so no money is coming soon.
- If the claim holds, it is the second Millennium Prize Problem solved with AI in three months, after an OpenAI model reportedly solved one in September.
- The claim is not yet confirmed, but the independent checks so far make it the most credible solution in the problem's history.
FAQs
What is the Riemann hypothesis?
The Riemann hypothesis, posed by Bernhard Riemann in 1859, says that all the interesting zeros of the Riemann zeta function lie on a single straight line, the critical line. It concerns the hidden pattern behind the distribution of prime numbers, and thousands of mathematical results depend on it being true. It is one of the seven Millennium Prize Problems, each worth $1 million.
Who claims to have proved it?
Levent Alpöge, a mathematician at Anthropic, announced on November 20, 2026, that he believes he has proved the Riemann hypothesis, and the generalised version, working with Anthropic's Claude AI model. He published the proof as code in Lean, a programming language that checks mathematical arguments step by step.
Has the proof been verified?
The Lean proof compiles without errors, and independent mathematicians, including Andrew Sutherland at MIT, have separately compiled and rechecked it, saying the chance of anything fishy is approximately zero. A final comparator run, which checks that the code does not redefine any terms or exploit a bug in Lean, is still pending. Full human understanding of the proof could take months or years.
Is there also a human proof?
Yes. Stanford professor Ravi Vakil posted a proof of the generalised Riemann hypothesis written by Stanford postdoctoral researcher Alessandro Masullo, noting that it was all human, with no AI involved. It is not yet clear whether the two proofs are related, and both need careful checking by the mathematics community.
What is the Millennium Prize?
The Clay Mathematics Institute offers $1 million for a proof of each of its seven Millennium Prize Problems, set in the year 2000. A prize is only awarded after a proof has been published in a refereed journal and survived two years of scrutiny, so no prize money is imminent even if the new claim is correct.
Why does this matter for AI?
If the claim holds, it would be the second Millennium Prize Problem solved with AI help in three months, after an OpenAI model reportedly solved one in September 2026. It would show that AI can do more than routine calculation, and can help solve the hardest open problems in mathematics, though it also raises questions about how humans verify and understand machine-produced proofs.
Was this article helpful?

Written By
TechorCoffeeContributing Writer
TechorCoffee is a technology reporter and software creator writing reviews and tutorials for TechOrCoffee.
Comments (0)
Leave a Comment
Read Next

Fri Oct 30 2026
Anthropic IPO: Claude Maker Files for Nasdaq Listing
Anthropic IPO explained: S-1 filing, ANT ticker, $16.3B revenue, $80.7B loss, $1.5-2T valuation talk, compute costs and what it means for Claude users.

Fri Oct 30 2026
Claude Opus 5: Same Price, Much Better Coding
Claude Opus 5 explained: July 2026 release, $5/$25 pricing, 1M-token context, Frontier-Bench and CursorBench results, and when to pick Opus, Sonnet or Fable.

Sat Nov 14 2026
Claude Fable 5.2: Anthropic's New Top-Ranked AI Model
Anthropic's Claude Fable 5.2 arrived on 9 November 2026 as the week's top-ranked new model. Here's what the launch means for users and developers.