It was late on a Sunday night when Levent Alpöge picked up his phone and typed a message to his boss. He had just watched an artificial intelligence do something no human had managed in 167 years — and he could not wait another day to tell the world.
Alpöge is a mathematician at Anthropic, the company behind the AI assistant Claude. For months he had worked with Claude on the most famous unsolved problem in mathematics. That Sunday night, the work came together — and on Friday morning, November 20, he published a four-page note on his personal website.
“I believe I have, working with Claude, discovered a proof of the Riemann Hypothesis,” he wrote. If he is right, an AI has just solved the problem mathematicians call the greatest mystery in their field — the day after Anthropic’s record-breaking stock market debut.
What the Riemann Hypothesis Actually Says
The Riemann hypothesis is a puzzle about prime numbers — 2, 3, 5, 7, 11 and the rest, the numbers that can only be divided by one and themselves. Primes are the building blocks of all whole numbers, yet they appear with no obvious pattern.
In 1859, the German mathematician Bernhard Riemann proposed a rule that governs where the primes fall. It is written in the language of the Riemann zeta function, a mathematical object that encodes the hidden order inside the primes.
Put simply, the hypothesis says that all the “nontrivial” zeros of that function sit on a single vertical line, where the real part equals one-half.
Computers have already checked the first 10 trillion zeros, and every one of them obeys the rule. But checking is not proving — and a proof has to cover infinitely many of them.
That is why the problem has survived 167 years of attacks by the best mathematicians on Earth. It is one of seven Millennium Prize Problems, carrying a $1 million bounty from the Clay Mathematics Institute. Until Friday, most experts believed a proof was still far away.
How Claude Cracked the Riemann Hypothesis
Alpöge did not simply ask a chatbot to solve the problem. He spent months working with Claude across many areas of mathematics, steering the effort toward a promising approach while the model supplied ideas of its own. The decisive stretch lasted 41 days, from October 6 to November 16.
The starting point was an older technique linked to the Russian mathematician Sergey Stepanov. Applying it to whole numbers had always run into a wall, because one part of the method had no obvious equivalent there. According to Alpöge, Claude found a way through by combining ideas from several branches of mathematics at once.
The breakthrough came through a long chain of transformations called theta lifts — a genuine mathematical idea that, by Alpöge’s own account, belonged to Claude. “Claude and I have had a great time together with such breadth in recent months,” he wrote, “but this time Claude far outpaced me near the finish line.”
It was not his first success with Claude. Earlier in the year, the model produced a counterexample to the 87-year-old Jacobian conjecture. In August, it raised the proven share of zeta zeros on the critical line from 41.6% to 67.2%.
A Proof No Human Has Read
Here is where the story turns strange. Alpöge has not published a conventional paper. Written out in ordinary mathematical prose, he says, the argument would run to more than 250,000 pages.
Instead, he released the proof as computer code — 17 million lines of it, written in Lean, a language built so software can check a proof line by line.
The files were split across 14 archive parts on his website, along with the four-page note. He also says the same method settles the Generalized Riemann Hypothesis, a broader statement covering a whole family of related functions.
And he is strikingly honest about what that means. Alpöge says he does not yet fully understand every argument Claude used. He plans to write the proof up properly, in human-readable form, once he can simplify it. For now, the world holds a masterpiece it cannot yet read.
Has the Riemann Hypothesis Proof Been Verified?
A proof written in Lean has one great advantage: a computer can check it. Independent mathematicians have already rebuilt Alpöge’s code, and it compiles without errors. MIT’s Andrew Sutherland said he verified the work on his own computer and knew of at least three other successful verifications.
Two separate proof-checking systems have also accepted the result, according to reports. That is a strong early signal — but not the finish line. A final layer of machine checking was still running when the news broke, and the Clay Institute still lists the problem as unsolved.
Even if every check passes, the formal process is slow. The Clay Institute requires publication in a qualifying journal, two years of general acceptance, and a dedicated committee review before the $1 million prize is awarded.
Leading number theorists say they think it will hold up. “The blast radius extends to the whole of mathematics,” said Ghaith Hiary of Ohio State University. “I didn’t think I’d see the day… but here we are.”
Others are more cautious. The excitement is real, they say, but so is the strangeness of a proof no human mind has fully grasped. As Maksym Radziwill of New York University put it, the point of mathematics is understanding why things happen, not just collecting statements of truth.
It is the same kind of machine checking Google DeepMind recently applied to the encryption inside Signal’s messaging app — but no machine has ever claimed a prize of this size before.
The Skeptics Question the Timing of the Riemann Hypothesis Claim
Not everyone is celebrating yet. Andrew Granville, a number theorist at the University of Montreal, called Alpöge’s four-page note “garbled” and said he should have taken the time to prepare a clearer write-up before going public.
The timing has also drawn attention. Alpöge announced the proof on Friday morning, the day after Anthropic’s record-breaking stock market debut.
The company had priced its shares at $83 and watched them close 17% higher at $97.17, briefly touching a $2 trillion valuation. Some wondered whether the announcement was meant to steady the share price.
Alpöge anticipated the question. “There is an obvious cynical view, to do with the timing, but it is false,” he wrote. He called the timing “horrible” — and said he simply could not live with waiting any longer. Anthropic itself declined to comment.
Why the Riemann Hypothesis Matters — Even If You Never Do the Math
Solving the Riemann hypothesis will not change your daily life tomorrow. But for mathematicians, it is the keystone of a vast building: hundreds of results in number theory are proved on the assumption that it is true.
A confirmed proof would turn all of those conditional results into certainties overnight. It would also deepen our understanding of the primes — and primes sit at the heart of modern cryptography, the mathematics that keeps online banking, messaging and shopping secure.
The deeper question is what this moment says about AI. In September, OpenAI’s models produced a proof of the quasi-Riemann hypothesis, a weaker version of the problem. A month later, the full claim arrived.
As one newspaper put it, humans alone solved one Millennium Prize Problem in 26 years; in three months, AI now appears to have done it twice.
That framing is exactly why this story matters beyond mathematics. If machines can now produce proofs humans cannot yet understand, the mathematician’s job is changing — from finding answers to checking, translating and truly understanding what the machine found.
What Happens Next for the Riemann Hypothesis Proof
The first task is verification. Mathematicians around the world are now running Alpöge’s code through their own machines, checking that the files build and the theorem really states the Riemann hypothesis. That will take weeks, not hours.
The second task is understanding. Alpöge has promised a readable write-up, and the community will want the argument in a form a human can follow. Until then, the proof sits in a strange middle ground: machine-checked, probably right — and still not understood.
The third task is the prize — and even in the best case, the Clay Institute’s two-year acceptance period means the $1 million will not be awarded any time soon. The problem will stay officially unsolved until the process runs its course.
For now, the claim stands as the most extraordinary announcement in mathematics in a generation — a glimpse of a future in which the hardest problems are solved first by machines and understood by humans only afterwards. Whether that future has truly arrived will be decided by the checking now under way, according to Science.



