"Working with Claude": Levent Alpöge's Riemann hypothesis proof passes Lean checks

On 20 November 2026 Anthropic mathematician Levent Alpöge released a proof of the Riemann hypothesis written in the Lean proof language. A day later the Lean and Mathlib communities certified the statement and the proof; a version written for mathematicians to read has not been published.

Graph of the real part (red) and the imaginary part (blue) of the Riemann zeta function along the critical line, from -30 to 30
Graph: Slonzor / Wikimedia Commons, public domain
In this material
  1. What the certificate covers
  2. What is still open
  3. A draft written without AI

The hypothesis, posed in 1859, says every non-trivial zero of the zeta function has real part 1/2; it governs how evenly the primes are spread. Alpöge, a number theorist who came to Anthropic from the Harvard Society of Fellows, announced the result in a four-page note on his site.

I believe I have, working with Claude, discovered a proof 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.

Levent AlpögeMathematician, Anthropicnote on alpo.ge, November 20, 2026

By his account he steered the argument and Claude did much of the technical work near the end, in fields he says he has not mastered. He published only the Lean code; a generated write-up ran past 250,000 pages.

What the certificate covers

The joint statement, endorsed by the Lean FRO and the Mathlib Initiative, reads: “We hereby certify that both the statement and the proof of the Riemann Hypothesis are correct.” The library depends directly only on Mathlib, Lean’s shared library of mathematics, and proves Mathlib’s own statement of the hypothesis. Two tools matched the theorem to that statement, and Lean’s kernel, the con-leche and con-ron checkers and independent checkers in other languages accepted the proof. The checks ran on an inert export, out of reach of any malicious code.

Andrew Sutherland of MIT, who published his own runs, notes the scope: the generalised hypothesis here is the one for Dirichlet L-functions, not for automorphic L-functions in general.

What is still open

The Lean statement looks forward to work “that will help explain the new ideas in this proof”. Brian Conrey, a specialist on the hypothesis, told AFP that “it looks like it’s seriously complicated” and that “there will be simpler ways to do it”. The $1 million prize of the Clay Mathematics Institute requires publication in a peer-reviewed journal and two years of scrutiny.

A draft written without AI

Vakil, a Stanford professor and president of the American Mathematical Society, posted the draft by Masullo, a postdoctoral researcher in his department. Brian Conrad had been examining it for months.

It is all human, with no AI. The hope was to have Brian work with Alessandro for a year to polish the writing and then announce it after running it by enough people.

Ravi VakilMathematician, Stanford University; president of the American Mathematical SocietyMath 216 blog, November 20, 2026

About eight hours later Vakil added that several people had reported problems in the draft, which runs past 900 pages, found with the help of AI models. He makes no claim as to the correctness of the proof strategy.

Alpöge writes that he will work to understand the argument, “to speak about it, to write it properly”, and expects shorter proofs to follow.

From the pulse

All news →
  1. Tools

    Thousands of passwords and tokens found in Claude Code settings files on GitHub

    Apiiro found database passwords, API tokens and root logins in .claude/settings.local.json files committed to public repositories. When a user approves a shell command with "Yes, and don't ask again", the command is saved in that file as written; Anthropic's documentation adds the file to the global .gitignore, which usually keeps it private.

    Source: arstechnica.com In the catalog: Claude

  2. Tools

    Anthropic steers Claude Code away from recreating existing games

    Anthropic confirmed that it added two sentences to the tool's instructions on the Opus 5.5 model telling it not to rebuild a specific existing game, even under a new name, because using Claude to infringe intellectual property breaks its usage policy; the X user who spotted the rule says it was pushed on 27 October. Staff member Addy Osmani said original games, genre games and games built from a user's own designs should keep working. It is unclear how the tool can tell whether a user owns an idea, and open-weight models have no such limit.

    Sources: gamedeveloper.com, gamesindustry.biz

  3. Business

    TSMC plans to invest up to $1 billion in OpenAI and Anthropic

    Its board approved up to $500 million for OpenAI preferred stock and up to $500 million for Anthropic shares at the expected price of its IPO, Reuters reported. Analysts quoted by The Wall Street Journal warn that such circular deals between suppliers and customers make real demand for AI harder to judge.

    Source: arstechnica.com

More in this section

All materials →
News

"Read by people": HD Moore caught AI providers keeping zero-retention prompts

From July to November 2026 HD Moore of Dragos sent about two million prompts with planted secrets to AI services. The secrets exposed breaches at three hosting providers and links from prompts marked for zero retention that were fetched days later.

News

Shinji Mikami wishes AI would "all just go away"; Kim Hyung-tae says games can't avoid it

At G-CON 2026 in Busan on 19 November, the director of the first Resident Evil and the head of Shift Up described how their studios use AI. Mikami keeps it away from design; Kim said game development without it no longer exists.

News

Cliff Bleszinski on Big Walk's AI backlash: "People really need to get over themselves"

House House said an AI chatbot helped with about 10 lines of Big Walk's code after IGF jurors refused to vote for games made with generative AI. Negative Steam reviews followed, and the Gears of War designer took the studio's side.