Claude May Have Solved the $1M Riemann Hypothesis in Six Weeks

By Hamza Rouane, the AI Whisperer · 2026-11-21 · 0 reads

Anthropic mathematician Levent Alpöge says he and Claude proved the 167-year-old Riemann hypothesis in about six weeks. Independent Lean checks have passed so far, critics call it garbled, and a human rival at Stanford just appeared.

An Anthropic mathematician says Claude helped him prove the Riemann hypothesis, the most famous unsolved problem in mathematics, and the computer-checked proof has so far survived every test outside experts have thrown at it. Levent Alpöge announced the claim on Friday, November 20, 2026, in a four-page note, saying he and a non-public version of Claude built the proof in about six weeks, according to the BBC.

The problem has stood since 1859. It carries a $1 million prize. And the timing could hardly be stranger: the note landed one day after Anthropic's stock market debut.

What happened, day by day

Alpöge is not a random poster. He is a number theorist who was a junior fellow in the Harvard Society of Fellows before joining Anthropic, and earlier this year he presented a counterexample to the Jacobian conjecture that he said was found with Claude, his Wikipedia page notes.

Here is how the Riemann claim unfolded, from the pages we read:

He did not release a normal paper. A full written version would run past 250,000 pages, so he released only the proof written in Lean, a programming language that lets a computer check every step, Futurism reported. Alpöge also says the method covers the much harder Generalized Riemann Hypothesis.

The question he could not dodge was the calendar. "The precise timing is horrible," he said. "There is an obvious cynical view, to do with the timing, but it is false," he added, per the BBC.

Why this one problem matters

Prime numbers look random. Riemann bet in 1859 that they hide a precise order, coded in the places where his zeta function equals zero. If every interesting zero sits on one vertical line, primes are spread about as evenly as they possibly can be, Nature explains.

That line has been checked for the first 10,000,000,000,000 zeros, the Clay Mathematics Institute says, as quoted by Tom's Hardware. Checking is not proving. Nobody had ever shown it holds for all of them.

Until now, the best AI-assisted result showed that at least 67.25% of those zeros follow the pattern. The new claim, if correct, says every single one does, without exception, mathematician Youness Lamzouri of the University of Lorraine told IBM Think.

The viral posts that set X on fire

The news broke on X before Anthropic said a word. Prediction market Polymarket posted it as "BREAKING" and the post passed 1.5 million views. In a reply, it put the chance that the hypothesis is officially declared solved by the end of the year at 26%.

Polymarket's breaking post on the Riemann claim, past 1.5 million views.

AI commentator Chubby (@kimmonismus) had posted about half an hour earlier, opening with "Oh my god." That post reached 609,500 views. A follow-up spelled out the part that stunned people: one mathematician, one model, six weeks.

Chubby's "Oh my god" post that reached 609,500 views.

Andrew Curran wrote that he had been waiting for an official Anthropic announcement, but the claim was "all over my timeline." His post drew 138,400 views, and later in the thread he shared a message from mathematician Bartosz Naskręcki saying Andrew Sutherland had finished the strictest checking run, and that the hypothesis was now "basically" 99.99999% true.

Andrew Curran's post, later updated with the independent Lean check.

The story spread far past X. Digg counted 3.8 million combined views across 189 sources in its first 17 hours, its story page shows.

What the experts are saying

The reaction from top mathematicians is split between awe and anger.

"If this fully checks out, it is clearly a historic milestone in mathematics, establishing one of the great mysteries of the subject." James Maynard, Oxford, Fields Medal winner, to IBM Think

Maynard added that he suspects the proof "will turn out to be correct," though even checking the Lean work will take time.

The strongest backing came from MIT's Andrew Sutherland, who compiled the proof independently. "Honestly the chance that anything is fishy here is approximately zero," he told Scientific American. Kevin Buzzard of Imperial College London pointed to the real trap: whether the theorem in the code is truly the Riemann hypothesis, and not something subtly different.

Lamzouri simply said: "I don't have words to describe what I am feeling right now. I am just in awe," per IBM Think.

Then came the knives. Andrew Granville of the University of Montreal said Alpöge should have written something clearer before announcing. "What he released is garbled," Granville told Nature, and he blamed the drop in Anthropic's stock for the timing.

Terence Tao was calmer: "I imagine the digestion process will take some time, though in this case at least we have a human author to communicate with," he said, per the BBC.

The twist: a human says they got there too

Hours after the Claude announcement, a second claim appeared. Ravi Vakil, president of the American Mathematical Society, wrote on his blog that a young Stanford mathematician, Alessandro Masullo, may have produced a proof months ago with no help from AI, Scientific American reports.

Stanford expert Brian Conrad, who has been giving feedback on that work, confirmed the blog post. But he said the paper is "undergoing major revision" and nobody has verified it. He called the decision to post it now a choice between "two unpalatable options."

So the oldest race in mathematics may end in a photo finish between a person and a machine.

The AI angle: how a chatbot ends up near a $1 million prize

This is the year AI started breaking math records. In September, OpenAI said staff using a non-public model with more than 10,000 AI agents solved part of a Navier-Stokes problem in 88 hours, and later it released claimed progress on 372 more open problems, the BBC reports. A group of scholars called this kind of corporate boasting "detrimental to the science of mathematics."

What makes the Riemann claim different is Lean. A language model can write confident nonsense. A Lean proof either compiles or it does not. That turns "trust me" into something any expert with a big computer can rerun, which is exactly what Sutherland did.

The catch is scale. The archive behind this proof holds about 175,000 Lean files, and its ideas draw on more than two centuries of mathematics, one researcher who mapped it wrote on X. Alpöge himself admitted he has not mastered all the math involved, AFP reported. The machine may have checked a truth that no human yet fully understands.

And no prize yet. Under the Clay institute's rules, a solution must be published in a peer-reviewed journal and survive two years of scrutiny, AFP notes.

What this means for you

Prime numbers sit under much of the math that keeps your bank card and your messages safe. A proof does not break that tomorrow. But it shows that AI tools can now push into the hardest thinking work humans do, in weeks, not decades.

If an AI can help close a 167-year-old problem, the jobs built on "only experts can do this" are next in line for the same shock. The skill that still matters is the one Lamzouri named: understanding why something is true, not just that it is.

Questions people ask

Has the Riemann hypothesis been solved?

Not officially. Levent Alpöge of Anthropic claimed a proof made with Claude on November 20, 2026, and independent checks of the Lean code have passed so far. Experts still need to confirm the code states the real hypothesis, and the Clay prize requires journal publication and two years of scrutiny.

Did Claude solve the Riemann hypothesis by itself?

No. Alpöge says he guided the work toward the right setting and that Claude "far outpaced" him near the finish. He used a non-public version of Claude over about six weeks.

Who is Alessandro Masullo?

A young Stanford mathematician who may have produced a proof without AI, according to a blog post by Ravi Vakil reported by Scientific American. That work is still under major revision and has not been verified.