Claude's Riemann Proof Is Certified Correct. No Human Understands It Yet

By Hamza Rouane, the AI Whisperer · 2026-11-21 · 2 reads

On November 21, 2026, the Lean FRO and Mathlib Initiative certified Anthropic mathematician Levent Alpöge's Claude-assisted, 17-million-line proof of the Riemann hypothesis as correct. Experts admit nobody yet knows why it works.

The Claude-assisted proof of the Riemann hypothesis, the most famous unsolved problem in mathematics, has been certified correct. On Saturday, November 21, 2026, the Lean FRO and the Mathlib Initiative said they have "full confidence" that the proof Anthropic mathematician Levent Alpöge released one day earlier, a 17-million-line Lean library built with Anthropic's AI Claude, is correct.

Here is the part that should make you sit up: the machines have checked it, but the humans have not read it. As one of the world's leading experts on computer-checked maths put it, "at the minute, we don't even know why the theorem is true."

What happened, day by day

1859. German mathematician Bernhard Riemann proposes a hypothesis about how prime numbers are spread along the number line. It becomes one of the seven Millennium Prize Problems picked by the Clay Mathematics Institute, each worth $1m, as The Guardian notes.

About six weeks before the announcement. Alpöge starts working on the problem with Claude. He told readers of his note that Claude helped develop the solution over "about six weeks", according to The Guardian.

Thursday, November 19. Anthropic goes public on the Nasdaq. Alpöge later admitted the clash: "The precise timing is horrible. If I could have, I'd have finished this all much earlier."

Friday, November 20. Alpöge posts a four-page explanation on his personal website. "I believe I have, working with Claude, discovered a proof of the Riemann Hypothesis," he wrote, Al Jazeera reports. He did not publish a normal paper. He released the proof as code in Lean, a programming language that checks every step of a proof. He also said the method covers the generalised Riemann hypothesis, a broader version of the problem.

Saturday, November 21. After a day in which the Lean and Mathlib communities "rallied to verify the proof's correctness", they publish a joint statement: "We hereby certify that both the statement and the proof of the Riemann Hypothesis are correct."

The posts that set the internet on fire

The news did not break through a journal. It broke on X.

On Friday afternoon, the AI account Chubby (@kimmonismus) posted the claim with the words "Oh my god" and a quote from Alpöge's note. The post passed 638,600 views.

The post that spread Alpöge's claim across X on November 20 (638.6K views).

A follow-up from the same account went deeper. According to that post, the first steps of the proof date to October 6, eight of its 20 steps landed on a single day, November 15, and the last entry, on November 16, simply reads "GRH." The post also says Alpöge admits he has little mastery of one area the proof depends on and credits Claude for those parts. That thread was seen 154,800 times.

The follow-up thread on the proof's six-week timeline and how much Claude did.

Then, on Saturday, came the post that turned a claim into a verdict. The official Lean account announced the certification and congratulated Alpöge. It reached 1.7 million views in a matter of hours.

The official Lean account announces the joint statement certifying the proof (1.7M views).

On Hacker News, the thread on Alpöge's claim collected 2,854 points and more than 2,000 comments within a day.

How a computer "certified" a proof no one has read

This is where it gets strange. Lean does not care who wrote a proof, a genius or a machine. It only checks that each step follows from the last.

But AI-written code can cheat. In a post on the Xena Project blog, published on November 12, days before the announcement, the author laid out four questions to ask before trusting any AI-generated Lean proof. Does the code state the right theorem? Does it compile? Does it prove the theorem as stated? Is the proof valid? It warned: "do not run random Lean code on your local machine unless you sandbox it first."

The joint statement says Alpöge's proof passes all four. The statement of the Riemann hypothesis it proves is the one already in Mathlib, Lean's human-checked maths library, "carefully vetted by experts in Lean and number theory". Dozens of people compiled the code and got the same result. Two independent tools confirmed the theorem proved is exactly Mathlib's version of the hypothesis.

And the proof itself was run through several separate checkers, including Lean's official kernel and checkers "implemented in different programming languages". Two of them come with a formal proof that they "cannot accept a false proof". The checks even ran on an inert copy of the proof, "out of reach of hypothetical malicious code".

Read that again. The experts built a quarantine around an AI's maths homework, in case the homework tried to trick the grader.

The reactions: awe, fear and wounded egos

Kevin Buzzard, professor of pure mathematics at Imperial College London, was blunt about how little humans grasp right now. "It's 17 million lines of Lean code, and this might correspond to tens of thousands of pages of mathematics," he told Al Jazeera. "At the minute, the proof is just millions of lines of gobbledygook written in a strange programming language."

Abhishek Saha, professor of mathematics at Queen Mary University of London, said the proof upended his expectations: "This claimed proof, if correct, shatters my model of reality." He also insisted the credit belongs to the human: "He should get a Fields Medal for this."

Computer scientist Scott Aaronson wrote on his blog that Lean has had kernel bugs in the past that "allowed proofs of 0=1". Still, he bet on the result: "I'm not 100% sure, but my bet is that the proof will stand." His 9-year-old son, he added, was "not even surprised at this point."

Not everyone is cheering. Warwick mathematician James Robinson had told The Guardian in September, after a rival AI claim by OpenAI, that it was "immature playground boasting writ large, underpinned by billions of dollars". And Buzzard named the feeling many in the field will not say out loud: "it has hurt a lot of people's egos."

There was a human rival too. On the same Friday, independent Italian mathematician Alessandro Masullo released an incomplete proof he said he had worked on for almost 12 years, without AI, because he "did not fully trust the system", per The Guardian, citing The New York Times.

What AI can already do here

This did not come out of nowhere. The Xena Project blog describes Anthropic formalising a proof of Fermat's Last Theorem in 13 million lines of Lean, and OpenAI's resolution of the Navier-Stokes problem in 640,000 lines. An earlier AI-assisted result, Al Jazeera notes, showed that at least 67.25 percent of the relevant values follow the pattern the hypothesis predicts. The new proof says all of them do.

The pattern is clear. AI models now produce proofs far longer than any human could write, and humans increasingly act as directors, checkers and translators. "Right now, as we speak, there's people all over the world who are translating parts of this programming language back into English," Buzzard said.

What this means for you

You may never think about prime numbers, but your bank does. Buzzard said some fast algorithms used in computing only work for sure if the generalised Riemann hypothesis is true, and cryptographers have long assumed it is. "We can sleep slightly safer in our beds at night because it is true," he said.

The bigger shift is about trust. A world-famous problem has been settled by a proof that no person has yet understood, and we believe it because other machines say so. That is how more and more of the work behind your money, your medicine and your phone may be checked from now on.

Questions people ask

Has the Riemann hypothesis been proven?

On November 21, 2026, the Lean FRO and Mathlib Initiative certified that Levent Alpöge's Lean proof of the Riemann hypothesis is correct. Human mathematicians are still working to understand and explain its ideas.

Did Claude solve the Riemann hypothesis on its own?

No. Alpöge, a mathematician at Anthropic, worked with Claude for about six weeks and wrote that "Claude far outpaced me near the finish line". Experts such as Abhishek Saha argue the main credit belongs to Alpöge.

Will Alpöge get the $1 million Millennium Prize?

The Riemann hypothesis is one of the Clay Mathematics Institute's seven Millennium Prize Problems, each with a $1m award. No prize decision has been reported yet; for now, the proof has a formal certification, not a prize.