SDSignal Desk

A Riemann proof claim lands in Lean, and the burden of proof shifts to the code

· Updated · Signal Desk newsroom

Analysis of reporting by The Verge. How we work

Image: The Verge

Anthropic mathematician Levent Alpöge says he and Claude proved the Riemann hypothesis and formalized it. The claim is unverified, but its machine-checkable form changes how fast the world can test it.

Key takeaways

  • Levent Alpöge, a mathematician at Anthropic, says he proved the Riemann hypothesis working with Claude; no independent verification exists yet.
  • The proof is presented in the Lean proof assistant and has been published only on his own website.
  • Anthropic has not commented or involved itself, and Alpöge suggests he waited until after the company’s Thursday IPO to publish.
  • The Clay Mathematics Institute requires publication plus at least two years of scrutiny and general acceptance before any $1 million prize.
  • The claim follows OpenAI’s contested Navier-Stokes result, a problem Alpöge had been working on with NYU’s Tristan Buckmaster.

The claim, stated plainly

Levent Alpöge, a mathematician who works at Anthropic, says he has proved the Riemann hypothesis with help from Claude. The problem dates to 1859, sits on the list of seven Millennium Prize problems, and concerns how the prime numbers are distributed through the zeros of a particular function. Few open questions in mathematics carry more weight; a great deal of number theory is written on the assumption that it is true.

Alpöge goes further than the headline problem. In a statement on his website he says the same method extends to the Generalized Riemann Hypothesis, and that he expects it to reach the Extended version with more time. He also says he and Claude formalized the arguments without assumptions. None of this has been independently checked, and The Verge, which first reported the claim, was careful to say so. Anthropic has stayed out of it entirely.

Why the Lean part matters more than the press cycle

Most extraordinary proof claims die slowly. Referees spend months, sometimes years, hunting for the gap. What makes this one different is that Alpöge presented it in Lean, a proof assistant that checks every logical step by machine. If the formal proof compiles, the question narrows sharply: not whether each line follows from the last, but whether the statement the code proves is really the Riemann hypothesis, and whether anything was smuggled in as an axiom. Those are questions specialists can answer in days rather than years.

That is the right way to make a claim this large, and it is a model other labs should copy. It also sets a trap for the excited reader. A clean compile on a subtly misstated theorem proves something else. Until experts in both analytic number theory and formal verification have read the statement line by line, the honest status is a serious, checkable claim, not a solved problem. The Clay Mathematics Institute’s own rules point the same way: publication, at least two years of scrutiny and general acceptance before the $1 million prize is even on the table.

The timing and the silence

The business context is impossible to ignore. Alpöge says he texted Anthropic chief executive Dario Amodei about the result late on a Sunday and indicates he held publication until after the company’s IPO on Thursday. That sequencing reads as an attempt to keep a personal mathematical claim from becoming a stock story, and Anthropic’s refusal to comment fits the same instinct. A newly public company has every reason to avoid touting a result it cannot yet stand behind. The pressures of that new life are exactly what we explored in our look at whether Anthropic can keep safety first as a public company.

There is also an awkward paper trail. In August, Anthropic said it was working on the Riemann hypothesis with Claude but did not expect the techniques involved to lead to a proof. Either the work moved far faster than the company expected, or something about the new claim will not survive review. Both are possible. The first would be one of the most remarkable jumps in AI-assisted research to date.

A rivalry written in proofs

This lands in a field already shaken by AI. Only one Millennium problem, the Poincaré conjecture, has been officially solved. In September OpenAI claimed Navier-Stokes after a large, costly push, and Alpöge was not a bystander: he had been working on that problem with NYU’s Tristan Buckmaster, who accused OpenAI of racing their route, a dispute we covered in the NYU mathematician’s account of how OpenAI handled it. OpenAI denies wrongdoing. It has since claimed more results, including the local smoothing and Exponential Time Hypothesis work, and stood up an advisory group of mathematicians to judge significance.

Alpöge, by The Verge’s account, also said in July that he had solved the Jacobian conjecture with Claude. Read together, the pattern is two labs, two styles. OpenAI runs institutional campaigns with huge compute and corporate announcements. Alpöge works as an individual researcher with a model as collaborator and publishes on his own page. The second model is closer to how mathematics has always worked, which may make it easier for the community to accept, provided the proof holds.

Who feels it

Mathematicians
A formal proof gives referees a concrete artifact to audit, but also adds to a backlog of AI-assisted claims the field is still digesting.
Anthropic
A verified result would be a landmark for Claude; a failed one would be awkward days after its IPO, which helps explain the company’s silence.
OpenAI
Its Navier-Stokes claim no longer stands alone as the headline AI math achievement, and the contrast in process will be noticed.
Cryptography and security teams
A proof would settle an assumption rather than break anything, so no immediate change is expected.

What to watch

  1. Independent experts confirming that the Lean statement faithfully encodes the Riemann hypothesis and uses no hidden axioms.
  2. Whether Alpöge submits the work to a journal, the first step toward the Clay institute’s two-year clock.
  3. Any statement from Anthropic, which has so far declined to comment on the claim.
  4. Responses from number theorists on the claimed extension to the Generalized Riemann Hypothesis.

FAQ

Has the Riemann hypothesis been solved?

Not officially. Levent Alpöge claims a proof developed with Claude, but it has not been independently verified, and the Clay Mathematics Institute requires publication, at least two years of scrutiny and general acceptance.

Why does the Lean formalization matter?

Lean checks each logical step by machine, so reviewers can focus on whether the formal statement matches the actual hypothesis and whether any assumptions were added, which can speed verification considerably.

Is Anthropic backing the claim?

No. Anthropic has not commented or involved itself. In August it said it did not expect Claude’s techniques at that time to lead to a proof.

Read the original

Continue at the source.

The Verge

Companies: OpenAI, Anthropic

More on OpenAI

All OpenAI coverage →