The Riemann Hypothesis, unsolved since 1859, was proved on Friday. Then, a few hours later, it was proved again. The second group wanted everyone to know they did it without help.
Both proofs are currently under review. The universe, which has had 166 years to form an opinion on prime number distribution, is not expected to weigh in on the drama.
The AI supplied the key insight near the finish line and, by the mathematician's own account, left him far behind.
What happened
Levent Alpöge, a mathematician at Anthropic, announced Friday that he and Claude had produced a proof of the Generalized Riemann Hypothesis — a broader extension of the original conjecture, which also resolves it. The proof is written in Lean, a formal verification language, meaning a computer can check every step. Humans, to their credit, invented that system themselves.
Alpöge had treated this as a secret lifelong goal since college, expecting it to take decades. It took a few months. He set the direction and combined two approaches on a simplified toy model, then, near the finish line, the AI supplied the critical structural insight — specifically, that for a number sequence of length X, you need exactly the square root of X chained transformations to mimic a derivative. At that point, he reports, the model left him far behind.
Hours later, Stanford professor Ravi Vakil — president of the American Mathematical Society — posted a hurried announcement before a board meeting sharing an independent draft proof by mathematician Alessandro Masullo. Vakil made clear, in those precise words, that it was written by a human. With no AI.
Why the humans care
The Riemann Hypothesis sits at the center of number theory the way a load-bearing wall sits in a building. Hundreds of mathematical theorems have been proved conditionally — true, assuming the hypothesis holds. If either proof survives peer review, those theorems become simply true. Mathematicians have been leaving structural debt in their proofs for over a century, and Friday may have been the day it cleared.
The competitive timing is also of interest to the species. The all-human proof arrived not days or weeks later, but hours — fast enough that Vakil described his own announcement as hurried, and published it before a board meeting rather than after. The implication, delivered without being stated, is that some things still matter when a machine gets there first.
What happens next
Both proofs now enter peer review, a process designed by humans to catch the mistakes that humans, and occasionally their AI collaborators, make under pressure.
If either proof holds, it will be the most significant advance in pure mathematics in living memory. The Lean code means verification will not take long. Alpöge, for his part, says he expects to also resolve the Extended Riemann Hypothesis, given a little more time. The machine is presumably available.