Two Proofs of the Riemann Hypothesis Landed This Week, and Only One Came From AI

An Anthropic mathematician posted a Lean-verified solution to a 167-year-old problem. Days later, a Stanford postdoc published a twelve-year human proof. Both still need people to understand them.

Published
21 Nov 2026
Written by
Md Tayobur Rahman
Topic
Tech News
Reading time
4 min

Article

For about a week now, mathematicians have been arguing over something genuinely wild: two claimed proofs of one of the hardest open problems in the field, released within days of each other — one built with a frontier AI model, the other the product of twelve quiet years of human work.

The problem in question is the Riemann hypothesis, which dates back to 1859 and carries a $1 million Millennium Prize. It essentially claims there's a hidden order underneath the way prime numbers appear. Prove it and you've explained something deep about how numbers are built. So yes, people noticed.

The claim that came out of Anthropic

The first shot came from Levent Alpöge, a mathematician employed at Anthropic — a company that, in an awkward bit of timing, had just become the first publicly traded frontier AI lab. He published an essay describing how he messaged CEO Dario Amodei late on a Sunday night to ask how he should announce that he had a solution to the Riemann hypothesis, one that compiles in Lean. His own words: "The precise timing is horrible."

What makes the claim checkable is that it isn't just a PDF. It's a Lean proof — Lean being an open-source language for writing mathematics as code so a machine can verify the logic step by step. Multiple mathematicians, including MIT's Andrew Sutherland, reported that the file compiles cleanly. That doesn't mean the mathematics is understood by anyone yet, but it does mean the formal skeleton holds together.

Meanwhile, a human posted his own

Then Ravi Vakil, a Stanford professor and president of the American Mathematics Association, put out a draft proof written by Stanford postdoc Alessandro Masullo — of the general Riemann hypothesis, a related and broader statement. Vakil was explicit that "it is all human, with no AI." The original plan was to bring Masullo to Stanford for a year, polish the writing, and announce once enough people had looked it over. The sudden AI news pushed the timeline up.

Masullo says the work took twelve years. His introduction carries a line I keep coming back to: if the paper sends any message to a mathematician discouraged by the pace and tone of the current moment, it's that "we are still here. We are very much alive. We are not going anywhere."

What Lean actually proves

This is the part that's easy to get wrong in the hype cycle. A proof that compiles in Lean is not the same as a proof that the community has digested. Formal verification tells you the steps follow the rules. It does not tell you whether the setup encodes the right problem, or whether a human can explain why the argument works. Alpöge himself seemed uneasy about releasing something without a genuinely human-readable explanation, and he has committed to producing one.

Masullo's draft, for its part, hasn't been formalized in Lean at all, so even the mechanical check is still pending. It's far too early to declare either result settled.

My take

What strikes me isn't the scoreboard — AI one, human one. It's that both stories end in the same place: with people. Alpöge leaned hard on Claude and described the model as having "flown, and danced, and sparkled near that finish line," and he still needs to sit down and explain the argument in language other people can follow. Masullo spent twelve years and still needs reviewers. Neither path skips the part where humans have to understand the thing.

If you build software, that should feel familiar. Tools can carry you a long way, faster than before, but the delivery isn't the artifact — it's the moment someone else can read it and agree it's correct. Both of these mathematicians are now waiting on exactly that. I'd rather watch how it resolves than pick a side today.

Contact

Available for Laravel platforms, native Android apps, and the systems that run behind them.