Anthropic mathematician Levent Alpöge announced on Friday that he and the company's AI model Claude have discovered a proof of the Riemann hypothesis — one of mathematics' seven Millennium Prize problems — and its broader Generalized Riemann Hypothesis. The claim, posted on Alpöge's personal website, arrives as a massive Lean formalization rather than a conventional paper. Independent verification has begun but remains incomplete.
What was announced #
Alpöge, a number theorist on Anthropic's technical staff, wrote: "I believe I have, working with Claude, discovered a proof¹ of the Riemann Hypothesis. In fact, the method easily extends much further, resolving the so-called Generalized Riemann Hypothesis, and I expect, with more time, the Extended Riemann Hypothesis too." [8]
Instead of a human-readable manuscript — which he said would exceed 250,000 pages — Alpöge released the argument as a Lean 4 project split into 14 compressed archive parts totaling roughly 239 megabytes and 175,187 source files. [2] Lean is a proof assistant that mechanically checks each logical step; if the code compiles against the standard mathematics library (Mathlib), the proof is considered logically valid relative to its definitions and axioms. [2][4]
Verification status #
Within hours, an independent verification effort confirmed that the downloaded archive matched Alpöge's mirror and that the top-level theorem statements compiled cleanly against Mathlib. The full automated replay of the entire proof was still running as of Friday. [2]
Mathematicians caution that three further checks are essential before the problem can be considered solved:
- Exact target: The Lean statement must match the standard Riemann hypothesis, not a subtly altered version.
- Faithful definitions: All mathematical objects in the code must accurately reflect accepted definitions.
- No hidden assumptions: The proof must not rely on unproven shortcuts or unintended axioms.[2]
Kevin Buzzard of Imperial College London summarized the mood: "Let's see if it compiles first," he said, reacting to the news. "Might be a while before we find the relevant experts who are able to tell us the thrust of the argument, assuming it's all true, of course, which it probably will be because of course Alpöge will have [a] huge amount of egg on [his] face if the Lean code doesn't compile or is deficient in some way." [6]
Context: a year of AI math accelerations #
The claim caps a year in which AI's mathematical output has moved from benchmark-chasing to frontier research. In July, Alpöge posted a Claude-generated disproof of the Jacobian conjecture. In August, Anthropic disclosed that an unreleased research version of Claude had raised the known lower bound for the proportion of zeta zeros on the critical line from 41.6% to 67.2% — progress on a related question, not a proof of the hypothesis itself. [9] In October, rival lab OpenAI released 722 AI-generated math manuscripts, including a claimed proof of a weaker "quasi-Riemann hypothesis." [1][2]
On the same day as the Riemann announcement, Alpöge and NYU mathematician Tristan Buckmaster also posted a separate 79-page Claude-assisted paper claiming a breakthrough on the unforced Navier–Stokes equations — another Millennium Prize problem — though its supporting numerical certificate has not yet been released. [2]
Timing and disclosures #
Anthropic's shares began trading on Nasdaq on November 19. Alpöge addressed the timing directly in his note, writing that the precise timing was horrible and that he would have finished much earlier if he could have. He said the result came together late Sunday night, that he messaged Anthropic CEO Dario Amodei (referred to by first name) about his ability to announce, and that there was a constraint to respect, though he asked about progressively more bizarre ideas. He added that the obvious cynical view about the timing was false. [8] Anthropic had not issued a public statement on the claim as of Friday morning US time. [1][4]
Alpöge also acknowledges unusual epistemic territory: "This puts us in the rather extraordinary situation this age has made possible, in which I am presenting the world with a complete argument before having mastered a massive neighbourhood around its mathematics." He adds that some technical descriptions paraphrase an AI-written summary that "might make zero sense," and that he personally has "little mastery of automorphic forms on higher rank groups," the area where Claude did much of the technical work. [8]
The mathematician and the method #
Alpöge earned his PhD at Princeton in 2020 under Fields Medalist Manjul Bhargava and served as a Junior Fellow at Harvard's Society of Fellows before joining Anthropic. [2][4] His note describes an approach adapting Stepanov's method — which proved the Riemann hypothesis for curves over finite fields — by building a chain of "theta lifts" whose length grows linearly with the sum length (X), while each theta lift contributes a square-root-like multiplicity, applied to von Mangoldt sums. [8] A rough timeline compiled by Claude dates the first steps to October 6 and the Generalized Riemann Hypothesis step to November 16. [8]
What happens next #
The Clay Mathematics Institute's Millennium Prize rules require a solution to be published in a qualifying peer-reviewed outlet, survive at least two years after publication, and achieve "general acceptance in the global mathematics community" before the $1 million prize can be awarded. The institute does not accept direct submissions. [2][4] As of Friday, Clay's official page still listed the Riemann hypothesis as unsolved. [2]
For now, the mathematical community faces a novel verification task: checking a 175,000-file Lean project whose human author says he has not yet mastered all of its mathematics. If the formalization holds — definitions faithful, axioms standard, proof complete — it would represent both a landmark in number theory and a striking instance of human-directed, AI-accelerated discovery. If it does not, it will join a long line of claimed proofs that collapsed under scrutiny. The Lean files are public; the checking has begun.
Sources #
- Anthropic mathematician claims to prove Riemann hypothesis, most ... (scientificamerican.com)
- Has the Riemann hypothesis been solved? Anthropic mathematician ... (clarin.com)
- Anthropic mathematician claims Riemann hypothesis proof with ... (superpowerdaily.com)
- Anthropic mathematician Levent Alpöge posts a claimed proof of the ... (mlq.ai)
- Anthropic employee claims general proof of the Riemann Hypothesis [pdf] (news.ycombinator.com)
- Anthropic mathematician claims AI-assisted proof of the Riemann ... (newscientist.com)
- Anthropic mathematician claims to have solved prestigious Millennium Prize problem with Claude | The Verge (theverge.com)
- https://alpo.ge/remark.pdf (alpo.ge)
- Claude has improved on a longstanding lower bound for the fraction of zeros of the Riemann zeta function that satisfy the Riemann hypothesis \ Anthropic (anthropic.com)