Key Points: Riemann Hypothesis Solved
- On Friday, November 20, 2026, Levent Alpöge, a mathematician at Anthropic, said he and Anthropic’s Claude AI had found a proof of the Riemann Hypothesis.
- A few hours later, Stanford postdoc Alessandro Masullo posted a proof of the Generalized Riemann Hypothesis (GRH) that he says is “all human, with no AI.”
- Neither claim has been confirmed. The Clay Mathematics Institute still lists the problem as unsolved.
- Alpöge’s proof is written in a computer checking language but is not readable by people. Masullo’s is readable but has not been formally checked.
- A $1 million prize is attached, but the Clay rules require a published paper and a two year wait.
Math rarely makes headlines. This week, it made two, within hours of each other.
Here is a plain guide to what was claimed, what has been checked and what nobody can say yet.
What Is the Riemann Hypothesis?
Prime numbers are the building blocks of math. They are numbers like 2, 3, 5 and 7 that divide only by 1 and themselves.
They look random as you count upward. In 1859, Bernhard Riemann found that their pattern is tied to a special function and to the spots where it equals zero.
The Riemann hypothesis says all the important zeros line up on one straight line. If that is true, the primes are spread out as evenly as possible.
Think of it as a promise about how closely you can predict prime counts. The problem is 167 years old and is one of seven Millennium Prize Problems.
Riemann Hypothesis Solved? What Levent Alpöge Claims
Alpöge, also spelled Alpoge, is the Anthropic mathematician at the center of the Anthropic Riemann Hypothesis story. He posted a four page note on his personal site.
Levent Alpoge’s Riemann note says Claude helped him reach the result, and that the AI “far outpaced” him near the end.
He released the work as Lean code, a language that lets software check each logical step. Reports describe it as millions of lines long.
His note also covers the generalized Riemann hypothesis for a broader family of functions. Reports on how long the work took differ, from several weeks to several months.
AP reports that Claude followed the code and judged it correct. That is a claim about Anthropic’s own tool, so independent checks matter more.
Alessandro Masullo and the Human Reply
A few hours later, Ravi Vakil, a Stanford professor and president of the American Mathematical Society, posted Masullo’s draft on his blog. He called it “all human, with no AI.”
Masullo says it is the result of 12 years of work. He wrote, “we are still here. We are very much alive.”
Vakil said the plan had been to bring Masullo to Stanford for a year of polishing. The competing news pushed him to release the draft early.
Stanford number theorist Brian Conrad has reviewed the draft privately for months. Vakil says it is “so far looking good,” but Masullo’s own paper warns it may contain errors.
What Is GRH, and Why Does It Matter?
The Generalized Riemann Hypothesis, or GRH, extends the same idea to a wider family of functions. It is a stronger and broader statement.
Many results in number theory and computer science are written as “if GRH is true, then.” A proof would turn a long list of conditional results into firm ones.
Reports differ on which claim covers what. Alpöge’s note is described as covering both versions, while Masullo’s is described as covering the generalized one.
Two Kinds of Evidence: Our Reading
Here is a point you will not find in most headlines. The two claims offer opposite kinds of proof.
Alpöge’s can be run through a computer checker, which is strict and fast. But a person cannot read it, so nobody yet understands why it works.
Masullo’s can be read by experts, who can follow the logic and judge its beauty. But it is an unpolished draft and has not been formalized.
One is checkable but opaque. The other is readable but unchecked. The strongest outcome would be a human proof that is also formalized.
That is our analysis, not a reported finding. But it shows why neither side can claim the question closed.
The Hidden Check: Did the Computer Prove the Right Thing?
A passing computer check is not the end. Reviewers must confirm that the statement written in Lean matches the real Riemann hypothesis.
If the definitions are slightly off, the computer could prove something different. One report calls this a separate step from checking the proof itself.
Early independent checks reportedly confirmed that the released files match and that the target statements can be written in Lean’s math library. The full proof check was still running on Friday.
What Experts Are Saying
James Maynard of Oxford, a Fields Medal winner, called it a potential historic milestone “if this fully checks out.” He said he suspects it will.
Alex Kontorovich of Rutgers said the strict machine checking gives strong reason for trust, but he has not yet understood it. Martin Bridson, president of the Clay Institute, said the path to real understanding will be long.
Steven Strogatz of Cornell wrote that “the plot thickens.” Sergei Gukov of Caltech said, “Now it’s getting more complicated.”
Five Questions to Ask Before You Believe a Math Breakthrough
- Has a journal published it? Clay requires a qualifying publication and two years of open review.
- Who has checked it? Look for named experts, not only the author.
- Is it machine checked, human checked or both?
- Does the checked statement match the real problem?
- Is the news from the authors, or from independent sources?
What Happens Next
Lean communities are replaying and auditing the files. Alpöge has said he will pursue a human readable version.
Mathematicians will read Masullo’s draft closely, and Conrad’s full review will carry weight. Credit and priority are unsettled.
For now, the honest answer to “Riemann hypothesis solved?” is simple. Maybe, and not yet confirmed.

