From a Lower Bound to the Riemann Hypothesis: Claude’s Month in Mathematics
Key points
- An unreleased Claude research model raised the proven lower bound on the share of Riemann zeta zeros lying on the critical line from 41.6% to 67.2%, using 31 million output tokens over two sessions 1.
- Anthropic mathematician Levent Alpöge claims that, working with Claude, he has proved the Riemann hypothesis and its generalized version, backed by a Lean formalization posted on his website 2 3 7.
- The claim is unverified: the Clay Mathematics Institute requires publication in a qualifying outlet, two years of scrutiny and general acceptance before any prize is considered 10.
- MIT mathematician Andrew Sutherland independently compiled the Lean result and called it a historic moment, while other voices urge rethinking how mathematical expertise is rewarded 3 5.
A Failed Attempt With a Surprise
The story begins with what Anthropic itself called an unreasonable challenge. A staff member, Jarred Sumner, who is not a mathematician, asked an unreleased research version of Claude to take a real stab at the Riemann hypothesis, the 1859 conjecture carrying a million-dollar bounty 1.
Claude did not prove it. But along the way it improved a longstanding lower bound on the fraction of zeta zeros known to lie on the critical line, lifting it from 41.6% to 67.2% 1. Anthropic said at the time that it did not expect the techniques to lead to a proof of the hypothesis itself 1.
Claude’s route combined recent work by Aryan and by Baluyot, Goldston, Suriajaya and Turnage-Butterbaugh, which frees Montgomery’s 1973 techniques from assuming the hypothesis, with a 2000 paper by Bombieri 1.


How Claude Worked
The process was heavily agentic. Claude first tried 650 ideas, none of which worked. After a nudge, it spent a day and a half coordinating about 60 subagents that ran 2,400 shell commands, wrote hundreds of Python scripts and checked results numerically against known zeros 1.
Subagents also refereed one another, searched for counterexamples, downloaded 54 arXiv papers to check novelty and re-proved the result from scratch. Anthropic mathematicians Levent Alpöge and Ralph Furman validated the paper, and Claude produced a Lean formalization with Eric Easley 1. Experts Brian Conrey and Dan Goldston also reviewed it on short notice 1.

The Claimed Proof
Months later, Alpöge announced on his website that, working with Claude, he believed he had proved the Riemann hypothesis, and that the method also resolves the Generalized Riemann Hypothesis 7. Reports say the announcement came the day after Anthropic’s IPO, and he says he told CEO Dario Amodei and weighed the timing carefully 2 3 7.
The submission is a short note plus a very large Lean archive, reportedly about 175,187 source files and 239MB 6. In his note, Alpöge describes a Stepanov-style argument adapted to the integers, using theta lifts, and credits Claude with surging ahead near the finish 6 7. He also concedes the argument is very long and likely to be simplified 7.
Sutherland independently compiled the Lean result and called the moment historic, though he said understanding the mathematics will take time 3. Hector Pasten likened a correct proof to a Rosetta Stone between branches of mathematics 3. Separately, early checks reportedly found that the files match the published archive and that the main statements compile against Mathlib, but researchers must still confirm the definitions and check for unsupported assumptions 6. The Verge noted that the solution has not been independently verified and that Anthropic had not commented 2.

Verification, Prizes and Context
Even a correct proof will not instantly earn a prize. Clay’s rules require publication in a qualifying outlet, at least two years since publication, and general acceptance in the global mathematics community 10. The Riemann hypothesis, formulated in 1859, is one of seven Millennium Prize Problems, each worth $1 million 9.
This is part of a pattern. Alpöge said in July that he had solved the Jacobian conjecture with Claude, and a post from him describes a proposed counterexample map 2 8. In a related and contested development, OpenAI claimed a Navier-Stokes solution amid allegations about its research practices, which it denies 2. Separately, Anthropic published claims that a Claude model made progress on two black hole puzzles, which Mashable noted arrived alongside the IPO 4.

What Comes Next for Mathematicians
The result has no immediate practical application, but Scientific American argues it would cap a year in which AI went from unreliable student to a superhuman force in the field 3.
Live Science argues that as production gets cheap, the scarce resource becomes human judgment about which questions and results matter. It calls for journals to demand clarity on what is new and what role AI played, and for institutions to stop treating publication counts as a proxy for quality 5. For now, the Riemann hypothesis remains officially unsolved while experts examine the proof 6.

Companies mentioned: Anthropic • OpenAI
Primary sources
Claude has improved on a longstanding lower bound for the fraction of zeros of the Riemann zeta function that satisfy the Riemann hypothesis (anthropic.com) – Anthropic's own announcement. It describes how an unreleased Claude, asked to attack the Riemann hypothesis, instead raised the proven fraction of zeta zeros on the critical line from 41.6% to 67.2%. It covers the background, the prior work it built on, the process (two Claude Code sessions, 31 million output tokens, 650 failed ideas, about 60 subagents), validation by Anthropic mathematicians and a Lean formalization. Published on anthropic.com, updated August 13, 2026.
Further sources
Anthropic mathematician claims to have solved prestigious Millennium Prize problem with Claude (theverge.com) – The Verge report on Alpöge's claim. It notes the lack of independent verification, Anthropic's silence, his message to Dario Amodei, Clay's prize rules, and the context of OpenAI's disputed Navier-Stokes claim and Alpöge's Jacobian conjecture work.
Anthropic mathematician claims to prove Riemann hypothesis, most famous problem in all of math (scientificamerican.com) – Scientific American's breaking news piece. It covers Alpöge's announcement and Lean proof, the generalized hypothesis, reactions from Hector Pasten, Carlo Pagano and Andrew Sutherland (who independently compiled the Lean result), and the broader significance for AI in mathematics.
Anthropic claims Claude solved 2 enduring mysteries about black holes (mashable.com) – A Mashable piece on a separate Anthropic claim that Claude Fable 5.2 made progress on two black hole problems (a phase transition in a small universe and a bound in a two-dimensional universe). It stresses that these have no direct application to our universe and coincided with the IPO.
An Anthropic mathematician claims to have proven the Riemann hypothesis. We have to rethink how we reward expertise. (livescience.com) – A Live Science Pro opinion article arguing that AI's flood of results means mathematics should reward judgment and significance over output. It recommends changes for journals, universities and training, and mentions OpenAI's unit distance and Navier-Stokes results.
Claude helped crack the 167-year-old Riemann hypothesis, Anthropic researcher claims (interestingengineering.com) – Interesting Engineering report by Aamir Khollam, dated November 20, 2026. It describes the four-page explanation and Lean archive (about 175,187 files, 239MB), the von Mangoldt sums and theta lifts, early checks of the files, remaining verification steps, and Clay's prize conditions.- Anthropic employee claims general proof of the Riemann Hypothesis [pdf] (alpo.ge) – Alpöge's own pdf statement. He claims a proof of the Riemann hypothesis and the Generalized version with Claude, with Lean formalization, describes a Stepanov-inspired argument with chained theta lifts, discusses the timing around Anthropic's IPO, thanks many colleagues, and includes a Claude-compiled timeline from October 2026.
Mathematician claims the Jacobian conjecture is false (x.com) – A short post on x.com in which Alpöge claims the Jacobian conjecture is false, giving an explicit polynomial map from C^3 to C^3 with Jacobian determinant -2 and three points mapped to the same value, and thanking Akhil and a collaborator named Fable.- The Millennium Prize Problems – Clay Mathematics Institute (claymath.org) – The Clay Mathematics Institute's Millennium Prize page. It explains that seven problems were chosen with a $1 million prize each, announced May 24, 2000 in Paris, and that the Riemann hypothesis dates to 1859 and also appeared on Hilbert's 1900 list.
- Rules for the Millennium Prize Problems – Clay Mathematics Institute (claymath.org) – The Clay Mathematics Institute's rules, adopted September 26, 2018. A proposed solution must be published in a Qualifying Outlet, wait at least two years, and win general acceptance in the global mathematics community; direct submissions are not accepted.

Powered by News Ranker