In short: in August 2026 an unreleased research version of Claude proved that more than two-thirds of the non-trivial zeros of the Riemann zeta function are simple and lie on the critical line. The previous record, set in 2020, was just over 41%. The result was checked by mathematicians, formally verified in Lean and posted on arXiv by Levent Alpöge and Ralph Furman.
What exactly was proved
The paper, More than two thirds of the zeta zeros are simple and on the critical line (arXiv:2608.13637), proves without any unproven assumptions that:
- at least 2/3 of the non-trivial zeros, counted with multiplicity, are simple and on the critical line (previous record: 5/12);
- at least 5/6 of them are distinct (previous record: 0.6603);
- with a refinement known as the Montgomery–Taylor window, those constants improve to 0.6725 and 0.8362.
The results extend to primitive Dirichlet L-functions, and the proofs are formally verified in Lean 4. Anthropic summarized the headline as a jump from 41.6% to 67.2%.
How Claude found it
Anthropic staff member Jarred Sumner, who is not a mathematician, asked Claude in Claude Code to take a real stab at the Riemann Hypothesis itself. In its first session Claude generated and tried about 650 ideas, and none worked. Sumner asked it to try again, sending little more than encouragement, and Claude spent a day and a half coordinating about 60 subagents. Together they ran some 2,400 shell commands, wrote hundreds of Python scripts, tested ideas against known zeros and refereed one another's work.
According to Anthropic, two of those subagents developed the key ideas, 13 contributed ideas to them, 13 acted as validators and two helped write the paper; the other 30 tried but failed to find new ideas. Claude downloaded 54 arXiv papers to check that the result was new, re-proved it from scratch, and recommended that a human number theorist validate it. The whole effort used 31 million output tokens.
Who checked it
Alpöge and Furman, both mathematicians at Anthropic, verified the argument and wrote an informal note for experts. Brian Conrey and Dan Goldston, two leading specialists on zeta zeros, examined the paper at short notice. Working with Anthropic's Eric Easley, Claude also produced a Lean formalization that passes comparator, a validation tool from the Lean community.
The idea, in brief
In 1973 Hugh Montgomery found a powerful way to study the spacing of zeta zeros, but his argument assumed the Riemann Hypothesis. Building on recent work by Aryan and by Baluyot, Goldston, Suriajaya and Turnage-Butterbaugh, together with a 2000 paper by Bombieri, Claude made that line of reasoning unconditional. It replaced the use of the Riemann Hypothesis with an inequality about the rank of a quadratic form built from Weil's Hermitian form, and used Sylvester's law of inertia to account for any zeros that might lie off the line.
Why 67% is not the Riemann Hypothesis
A proportion result says that, counted over larger and larger ranges, at least two-thirds of the zeros are on the line. It still leaves room for infinitely many exceptions. The Riemann Hypothesis demands that every non-trivial zero, without exception, lies on the line. Anthropic said at the time that it did not expect these techniques to lead to a proof of the Riemann Hypothesis.
A human follow-up
Youness Lamzouri of the University of Lorraine found the original paper hard to digest, so he went back to the papers Claude had drawn on. He spotted a simplification and produced a shorter proof of the same 67.25% figure, along with a new bound: at most 11% of zeros can be neither simple nor on the line. Oxford's James Maynard, a 2022 Fields Medal winner, told Live Science that the Claude proof contains new ideas that engage more directly with the problem, and described Lamzouri's reframing as a positive example of how mathematics can move forward with AI.
From 67% to a claimed proof
Three months later, Alpöge announced a claimed proof of the full Riemann Hypothesis and GRH with Claude, by a different route rooted in Stepanov's finite-field method. Read the overview of the claimed proof made with Claude, check the verification status, or place both results in the Riemann Hypothesis timeline.