Knowledge work · 21 Nov 2026

Every zero. Who checks the proof?

A mathematician and Claude announced a formal argument about the Riemann hypothesis. A proof checker can test the artifact independently of the model. The harder human task is establishing that the statement and assumptions mean what the authors claim.

Riemann asked the question in 1859. On Friday, MIT mathematician Andrew Sutherland ran the finished argument through an independent proof checker. The verdict came back in 3 minutes, 15 seconds.

The question was whether every nontrivial zero of the zeta function lies on one vertical line. It matters because those zeros tell us how far the count of prime numbers can stray from its predicted value. Mathematicians had already checked more than 12 trillion zeros. That still left infinitely many opportunities for an exception.

In August, Levent Alpöge helped check Claude’s result that at least 67.2% lie on the line. By November, the same mathematician and Claude had produced an argument for every one. He chose the starting point. He credits Claude with working out how many transformations, called theta lifts, to chain together in a field he says he has yet to master.

Two independently written proof checkers accepted the full export. A separate check matched the result to the Riemann Hypothesis as mathematicians had already defined it.

That same day, Alessandro Masullo released his own 940-page draft, nearly twelve years in the making. Within eight hours, mathematicians using several AI models had reported problems in it.

Switch Claude off tomorrow and we can still check the proof. The mathematician who helped find it is still learning how to explain it. That’s the part I can’t get past.

← Back to the archive