Anthropic's Claude Claims Riemann Proof; Stanford Counters All-Human

Anthropic's Levent Alpöge says Claude proved the Riemann hypothesis in Lean; hours later Stanford released an independent all-human proof draft.

@sharedot

Anthropic's Claude Claims Riemann Proof; Stanford Counters All-Human

By @sharedot · · 6 pages

  • Programming
  • AI
  • Mathematics
  • Riemann Hypothesis

Anthropic's Levent Alpöge says Claude proved the Riemann hypothesis in Lean; hours later Stanford released an independent all-human proof draft.

Anthropic claims an AI-assisted proof of math's hardest problem

Anthropic mathematician Levent Alpöge announced Friday that he believes he has, working with Claude, discovered a proof of the Riemann hypothesis — and, going beyond the original 1859 prime-number problem, the broader Generalized Riemann Hypothesis.

According to SBS News, Alpöge released a four-page document and a version of the proof written in Lean, the programming language that lets computers verify mathematical proofs step by step against math's basic rules.

The announcement came just one day after Anthropic's record-breaking IPO, timing Alpöge himself called potentially cynical-looking. He says he made the breakthrough late Sunday night, then consulted CEO Dario Amodei before publishing, arguing waiting longer simply was not an option.

Why the proof looks like 250,000 pages of Lean, not prose

The evidence is unusual: rather than a readable paper, Alpöge released the proof only as Lean code split into 14 data packages.

Printed out, the argument would run over 250,000 pages, because the AI spells out even trivial facts such as a number other than one-half being either larger or smaller than one-half. Alpöge told outlets that publishing such unreadable AI 'slop' would be inappropriate.

The Decoder reports the key insight came from Claude: for a number sequence of length X, you must chain exactly the square root of X special transformations together to mimic a derivative on whole numbers, extending an idea from mathematician Stepanov.

Alpöge admits Claude 'flew far ahead' of him and that he presented a result whose surrounding math he has not fully mastered.

Hours later, Stanford answers with an all-human proof draft

In a dramatic counter just hours after Anthropic's announcement, The Decoder reports that Stanford math professor and American Mathematical Society president Ravi Vakil posted a 'hurried announcement' sharing an independent draft proof of the Generalized Riemann Hypothesis by mathematician Alessandro Masullo — 'all human, with no AI,' Vakil wrote.

Renowned number theorist Brian Conrad had quietly reviewed Masullo's work for months, and the draft was meant to be refined for a year before publication; Anthropic's sudden announcement forced the researchers to release early.

Within eight hours, Vakil updated his post saying several people, some aided by AI models, had already flagged problems, and he made no claim as to the correctness of the proof strategy.

What a verified proof would change

Neither claim has been peer-reviewed yet, and verification will take time: the Clay Mathematics Institute requires peer-reviewed journal publication and a two-year verification period for its $1 million Millennium Prize.

SBS News notes hundreds of papers assume the Riemann hypothesis is true, so a confirmed proof would turn those into established theorems overnight, with repercussions across cryptography, quantum physics and computer science, since primes underpin RSA encryption.

Both draft proofs cover the Generalized Riemann Hypothesis, extending the stakes to a large family of related equations. Alpöge plans a human-readable version over coming months, while Vakil urged mathematicians to translate Masullo's draft into Lean and close small gaps.

If either holds up, 2026 will be remembered as the year math's north star fell.

Sources

  1. news.sbs.co.kr › Anthropic's Claude Claims to Prove Riemann Hypothesis, One of Mathematics' Greatest Unsolved Problems
  2. the-decoder.com › Hours after Anthropic claims a Claude proof of the Riemann hypothesis, Stanford releases an all-human competing draft

More on Programming