Claude Helped Crack the Riemann Hypothesis — AI Solves Maths' 167-Year-Old Greatest Mystery
On Friday 20 November 2026, a mathematician at Anthropic announced that he believes he has solved the most famous unsolved problem in mathematics — with the help of an AI. Levent Alpöge says he worked with Anthropic's Cl
On Friday 20 November 2026, a mathematician at Anthropic announced that he believes he has solved the most famous unsolved problem in mathematics — with the help of an AI. Levent Alpöge says he worked with Anthropic's Claude to produce a proof of the Riemann hypothesis, a puzzle about prime numbers that has resisted every attempt since 1859. Within a day, independent mathematicians had checked the proof with a computer and confirmed it holds. If it stands, it is the biggest result in the history of AI-assisted mathematics — and one of the biggest moments in mathematics itself.
What is the Riemann hypothesis?
First posed by the German mathematician Bernhard Riemann in 1859, the hypothesis is about the distribution of prime numbers — the numbers divisible only by themselves and 1. It proposes a hidden pattern in how primes are spaced out, encoded in a mathematical object called the Riemann zeta function. Proving it would confirm that the primes follow a precise order rather than appearing at random.
The problem is one of the seven Millennium Prize Problems set by the Clay Mathematics Institute, each carrying a $1 million prize. It is widely regarded as the most important open question in pure mathematics, and it has defeated every mathematician who has tried it for 167 years.
How the proof was found — and checked
Alpöge, who works at Anthropic, said he arrived at the solution on Sunday night and texted the company's chief executive, Dario Amodei, to discuss how to announce it. Rather than publishing a conventional paper, he released the argument in Lean — a programming language that lets mathematical reasoning be checked step by step by computer.
That choice is what made the rapid verification possible. If a Lean proof compiles without errors, the logic of every step is confirmed. Independent researchers, including MIT mathematician Andrew Sutherland, separately compiled Alpöge's proof and reported no issues. The Lean and Mathlib communities then issued a joint statement: "We have full confidence that Alpöge's proof of the Riemann Hypothesis is correct, and we congratulate him on this achievement." Sutherland put it more bluntly: "Honestly the chance that anything is fishy here is approximately zero."
Alpöge described the work as a "stream-of-consciousness jaunt" through the problem with Claude, writing that the model "flew, and danced, and sparkled near that finish line." He said the method also resolves the generalised Riemann hypothesis, a broader version of the same problem covering a whole family of related functions.
A human proof arrived the same day
In an extraordinary coincidence, a second proof appeared on the same day — this one entirely human. Ravi Vakil, a Stanford professor and president of the American Mathematical Society, posted a proof of the General Riemann hypothesis written by Stanford postdoctoral researcher Alessandro Masullo, noting that "it is all human, with no AI." That proof has not yet been expressed in Lean, so it is too early to say whether it checks out.
Alpöge and a collaborator, NYU mathematician Tristan Buckmaster, also released another important proof related to the Navier-Stokes equations — the problem that kicked off the race between frontier AI labs and mathematicians this summer, when OpenAI said one of its models had solved a related Millennium Prize problem on the equations that govern how fluids move.
The scepticism — and the timing question
Not everyone is comfortable with how the announcement was handled. Number theorist Andrew Granville of the University of Montreal called the four-page document Alpöge released "garbled" and said he should have prepared a clearer write-up before announcing. Others noted the timing: the announcement came the day after Anthropic's record-breaking stock market debut, when its shares were already falling. Alpöge addressed that directly in his note, writing that "there is an obvious cynical view, to do with the timing, but it is false."
The scepticism is about presentation, not the mathematics. A Lean proof either compiles or it does not, and this one compiled — independently, more than once. What is still missing is a human-readable explanation: the mathematical community is still working through what the proof actually says, and the Clay Mathematics Institute has not yet ruled on the $1 million prize.
Why this matters beyond mathematics
- AI is now doing frontier research, not just assisting it. In August, Anthropic said an unreleased version of Claude had improved a long-standing result related to the Riemann hypothesis — while stressing it had not proved it. Three months later, the full proof arrived.
- Formal verification changes the game. Because the proof is written in Lean, it could be checked by anyone with a computer in hours, not the years a traditional peer review of a result this size would take.
- The race is now between labs and mathematicians. OpenAI's fluid-dynamics result, the human proof from Stanford, and now Claude's Riemann proof all landed within months of each other. The next Millennium problem may fall just as fast.
Bottom line
A 167-year-old problem that defeated every human mathematician was solved by a human working with an AI — and a computer has already confirmed the logic is sound. The proof still needs to be understood by humans, and the prize still needs to be awarded, but the headline is already historic: AI has crossed from helping with mathematics to making mathematical history. Watch for the Clay Institute's response, and for a readable explanation of the proof in the weeks ahead.
Sources: TechCrunch — AI claimed a solution to the 167-year-old Riemann hypothesis · Scientific American — The AI Riemann hypothesis proof is true, independent mathematicians confirm · Nature — Mathematician says he and AI solved Riemann hypothesis. Hero photo: Bernhard Riemann, c. 1863, public domain, via Wikimedia Commons.
Originally published by Dev.to AI. Aggregated on AIWithGhost for educational purposes — full credit and traffic to the original publisher.