TechyCluster AI Can anyone check the Riemann hypothesis Claude proof?

Can anyone check the Riemann hypothesis Claude proof?


Can anyone check the Riemann hypothesis Claude proof?

Can anyone check the Riemann hypothesis Claude proof?
Image: TechyCluster

Key takeaways

  • Anthropic mathematician Levent Alpöge says he proved the Riemann hypothesis with Claude, but the claim sits on his personal website and has not been independently verified.
  • The proof is written in Lean, a language that lets software check every logical step, though the check only means something if the Lean statement matches the real hypothesis.
  • The Clay Mathematics Institute wants a solution in print and tested for at least two years before it will consider a Millennium Prize, and it has not publicly accepted this claim.
  • According to TechRadar, there is currently no human-readable version of the proof, which makes ordinary peer review hard to do.

Levent Alpöge, a mathematician at Anthropic, says he has proved the Riemann hypothesis with help from Claude. He posted the claim on his personal website the same week as Anthropic’s Thursday IPO. No one outside has confirmed it, and the 167-year-old problem has a reputation for eating confident announcements.

The question most readers will have is simple: how would anyone know if it’s true? There are two checks, and they work very differently.

What has Alpöge actually claimed?

Alpöge says he and Claude produced a proof of the Riemann hypothesis, a problem proposed in 1859 about the distribution of prime numbers and the zeroes of a function. TechRadar reports that he believes the two of them have found a proof.

He also says the method settles the Generalized Riemann Hypothesis, and he expects it to handle the Extended Riemann Hypothesis given more time. He credits Claude with the final breakthrough and says it “far outpaced” him near the finish line. The Verge reports he also claimed in July to have solved the Jacobian conjecture with Claude.

The claim is a personal-website post, not a journal paper. That matters for everything below.

How does a Lean proof get checked?

Lean is an open-source proof language, and it’s also a programming language. You write every step of an argument in strict code, and the software rejects any step that doesn’t follow from what came before. If the whole file is accepted, the logic holds. No referee has to trust the author’s reasoning.

Alpöge says he and Claude formalized the arguments without assumptions, according to The Verge. That’s the property that makes a Lean proof attractive. A human referee can miss an error buried deep in a very long paper. A proof checker doesn’t get tired.

It also matters because of size. TechRadar says an early draft apparently ran past 250,000 pages, and that Alpöge didn’t want to publish an unreadable mess. No human is going to referee that line by line. A machine can.

What Lean can’t tell you

A passing Lean file proves exactly what the file says it proves, nothing more. If the statement at the top is subtly different from the real Riemann hypothesis, because a definition is off or an assumption is tucked away, the software will happily certify the wrong thing.

So the first job for outside experts isn’t re-running the proof. It’s reading the statement and the definitions it depends on, and asking whether they match the problem mathematicians have worked on since 1859. That is one of the open questions around this claim, and a small one compared with the proof itself. A short statement is far easier for humans to audit than a giant argument.

There’s a second gap. TechRadar says there is currently no human-readable version. A verified result that nobody can read would settle whether the hypothesis is true, but it wouldn’t teach mathematicians why. Most of them will want the why.

Laptop screen showing lines of code in an editor
Photo: Autopilot via Wikimedia Commons, CC BY-SA 4.0

Why Clay’s two-year rule still applies

The Clay Mathematics Institute runs the Millennium Prize. Before it considers an award, a solution has to appear in print and survive at least two years of examination, and mathematicians worldwide must broadly accept it. The Riemann hypothesis is one of seven problems, each worth $1 million. Clay has not publicly accepted this claim.

Lean doesn’t shortcut that rule as far as anyone has reported. As I read it, a website post isn’t publication in the sense the rule means, so the clock hasn’t clearly started. And the rule is about acceptance by people, not just a green checkmark from software.

The track record explains the caution. The Poincaré conjecture is the only Millennium problem officially settled since the prize launched about 25 years ago. OpenAI claimed in September to have cracked Navier-Stokes. The Verge calls that result widely considered credible, yet it has no formal recognition, and it has been dogged by allegations, which OpenAI strongly denies, that it used unpublished work mathematicians shared with its AI tools.

Should you believe it yet?

Not yet, and not because the claim is silly. In August Anthropic said it had Claude working on the Riemann hypothesis, yet it didn’t expect the techniques in use to produce a proof. Alpöge’s claim arrives after that, in a field where AI-assisted results are piling up faster than people can absorb them.

Timing also invites suspicion. Alpöge says he messaged Dario Amodei, Anthropic’s CEO, late Sunday night, and his document suggests he held off publishing until after Thursday’s IPO. He calls the timing “horrible” and acknowledges an obvious cynical reading tied to it, which he insists is false. The Verge says Anthropic hadn’t responded to a request for comment at publication, and the company hasn’t involved itself in the announcement.

For a reader without a math degree, the practical test is who checks it and when. Watch for three things: Lean experts confirming the statement is the real one, mathematicians publicly working through the argument, and a readable version. A claim that clears all three would be a very different story from a post on a personal site.

If you want to understand what’s at stake while you wait, John Derbyshire’s Prime Obsession is a popular book-length account of the Riemann hypothesis, written for people who aren’t mathematicians.

Mathematicians discussing equations in a university lecture hall
Photo: Tungsten via Wikimedia Commons, Public domain

What nobody knows yet

The big one is whether the proof is correct. Beyond that, the gaps are specific:

  • Whether independent mathematicians or the Lean community will verify the formalization, and whether it states the hypothesis faithfully.
  • Whether a human-readable version will come out, and when.
  • How the work divides between Alpöge and Claude. Which model and setup he used hasn’t been reported.
  • Whether Anthropic will endorse or review the claim, given that Alpöge works there.
  • How the result squares with the company’s August doubts.
  • What the Clay Institute will say, and whether the work goes through a journal.

Anthropic’s separate blog post this week about Claude Fable 5.2 and two black-hole puzzles is a different story. According to Mashable, Anthropic itself says those findings apply only to simplified theoretical universes and have no practical use. It shouldn’t be read as support for the Riemann claim. For the business backdrop to all of this, see our look at the Anthropic IPO filing.

Tools & products mentioned

Disclosure: some links in this article are affiliate links or our editor's own courses and book. If you buy through them, TechyCluster or its editor may earn money at no extra cost to you. This does not influence what we cover or how we cover it. As an Amazon Associate, TechyCluster earns from qualifying purchases.

Frequently asked questions

Has the Riemann hypothesis been proved?

Not in any established sense. Levent Alpöge, a mathematician at Anthropic, says he proved it with Claude and posted it on his personal website. The proof has not been independently verified.

What is Lean and why does it matter here?

Lean is an open-source proof language that lets software check each logical step of an argument. Alpöge presents his proof in Lean. A passing check is only meaningful if the Lean statement faithfully captures the real Riemann hypothesis.

Will Alpöge win the $1 million Millennium Prize?

Not soon, if at all. Clay wants a solution in print, tested for at least two years, and broadly accepted by mathematicians before it considers an award. It hasn't publicly accepted this claim.

Did Anthropic announce the proof?

No. The Verge reports Anthropic hasn't commented on or involved itself in the announcement, and it hadn't answered a request for comment at publication. The claim came from Alpöge's personal website.

Get the next one in your inbox. One email a day with the tech stories that matter, explained in plain English. Subscribe free.

Sources

This article was compiled from reporting by the following outlets. Links go to the original reports.

  1. The Verge: Anthropic mathematician claims to have solved prestigious Millennium Prize problem with Claude
  2. TechRadar: Anthropic mathematician says Claude helped him solve a 167-year-old math problem
  3. Mashable: Anthropic claims Claude solved 2 enduring mysteries about black holes

About the author
, Aerospace engineer & author

Sandeep Bandyopadhyay is a mechanical and aerospace engineer with more than 20 years in aircraft structures and composites, including work on Boeing's 777-9 wing and GE Aviation nacelle programs. He holds an MBA and a PMP, and is the author of "Aerospace Structures and Composite Materials with Artificial Intelligence" (2025). At TechyCluster he writes about AI, hardware and the technology decisions that affect engineers, businesses and everyday users.

Reported and edited by Sandeep Bandyopadhyay, with AI research tools. Sources are listed below; corrections are welcome at editor@techycluster.com.

Leave a Reply

Your email address will not be published. Required fields are marked *