The Daily Inference
Science & Space · News · Unconfirmed

Claim: AI helped crack the Riemann hypothesis, 167 years on

Levent Alpöge says he and Claude have solved a problem that has stood since 1859. An independent software check is an important step, but mathematicians still need to establish exactly what the argument proves.

Unconfirmed: this story rests on reports that no primary source has confirmed yet. Later editions will follow it up.

Anthropic mathematician Levent Alpöge says he and Claude have proved the Riemann hypothesis, a 167-year-old problem whose resolution would reshape research into the distribution of prime numbers. [2] [3]

Alpöge announced his claim on November 20, 2026, and released an argument written in Lean, a language that allows software to check the logical steps of a mathematical proof. He also said the method proves the generalized Riemann hypothesis, a broader mathematical conjecture. [2] [3]

There is already one significant independent check. Mathematician Andrew Sutherland compiled Alpöge's result in Lean, according to Scientific American.

But Sutherland said substantial work remains to understand its mathematical details. The claim is unconfirmed, and the Clay Mathematics Institute still lists the hypothesis as unsolved. [2] [4]

The immediate task is to establish that the software has checked the theorem mathematicians actually want proved, using sound definitions and assumptions. Then researchers need to understand how the argument works. A successful software check is powerful evidence, but those are separate questions. [2] [3]

No human-readable version had been published, according to New Scientist. Alpöge said an unpublished version ran to more than 250,000 pages. That makes the next stage an unusually demanding exercise in mathematical inspection. [3]

What has supposedly been solved

Bernhard Riemann formulated the hypothesis in 1859. It concerns the zeta function, a mathematical object closely connected to the distribution of primes, the numbers greater than one that divide evenly only by themselves and one. [4]

Primes have a pattern in the aggregate, even though their individual appearances are irregular. The hypothesis addresses how far their actual distribution can stray from its average pattern. That is why a statement about a function's zeros has become central to work on whole numbers. [4]

A zero is an input at which a function produces zero. The relevant inputs here are complex numbers, which have two components, a real part and an imaginary part. They can be drawn as points on a plane, rather than placed along a single number line.

The Riemann hypothesis says every non-obvious zero of the zeta function has a real part of one-half. On that plane, they would all sit on the same vertical line, known as the critical line. [4]

The Clay Mathematics Institute says the first 10 trillion relevant zeros have been checked. That is an enormous body of numerical evidence, but it cannot establish a statement about every zero. A proof must rule out an exception anywhere, including beyond the reach of those calculations. [4]

If Alpöge's argument holds, results that currently assume the hypothesis could become unconditional theorems. Researchers could use them without first attaching the mathematical equivalent of an asterisk. The significance would reach well beyond the single problem. [2]

That does not mean an immediate technological windfall. Scientific American said a correct proof would have no immediate real-world application such as changing global banking or making the internet faster. Its first consequences would be inside mathematics. [2]

What Lean can establish

In a conventional proof, mathematicians read an argument and judge whether each step follows. In Lean, definitions, assumptions and deductions are expressed in a precise language so that software can check their logical connections. Alpöge has released his proposed proof in that form. [2] [3]

This offers a way to inspect an argument too large for ordinary line-by-line reading. A formal checker can test chains of deductions without relying on a reader to notice every missing step. But it checks the statement it has been given, under the assumptions it has been given.

For this claim, reviewers must therefore examine whether the encoded statement precisely matches the standard Riemann hypothesis. They must also check the definitions and assumptions on which the deduction rests. New Scientist said the code and mathematics still required thorough human checking. [3]

Sutherland's independent compilation gives the announcement more substance than a declaration alone. His warning about the work ahead also identifies the next obstacle: turning a machine-checked argument into mathematics that other specialists can understand and assess. [2]

The size of the unpublished account makes that task formidable. A readable explanation would need to expose the central ideas, show where the decisive step occurs and allow experts to connect those ideas to the formal files. Length is no measure of correctness, and 250,000 pages is no substitute for a clear route through an argument. [3]

A much bigger leap than August

Anthropic announced related progress in August 2026. It said Claude had improved a known lower bound for the proportion of zeta zeros on the critical line from 41.6 percent to 67.2 percent, an increase of 25.6 percentage points. [1]

That result concerned a guaranteed proportion of zeros. The Riemann hypothesis concerns all of them. Anthropic explicitly said it did not expect the techniques behind its August work to prove the hypothesis. The earlier result provides context for the new claim, not confirmation of it. [1]

Alpöge's announcement goes further still. He said the method resolves the generalized Riemann hypothesis and that, with more time, he expects it could extend to the Extended Riemann Hypothesis. Those are additional claims requiring their own scrutiny, rather than consequences that can yet be treated as established. [2]

The next useful advance is an independent mathematical assessment of the released formalization, alongside an explanation researchers can read. Sutherland has completed a software check. The work he says remains is to understand the mathematics it contains. [2]

Topics: Anthropic · Claude · Mathematics

Every edition in brief, three times a day, on our Telegram channel, on Bluesky and on Threads.

Sources
  1. Claude has improved on a longstanding lower bound for the fraction of zeros of the Riemann zeta function that satisfy the Riemann hypothesis \ Anthropic anthropic.com
  2. Anthropic mathematician claims to prove Riemann hypothesis, most famous problem in all of math | Scientific American scientificamerican.com
  3. Anthropic mathematician claims AI-assisted proof of the Riemann hypothesis | New Scientist newscientist.com
  4. Riemann Hypothesis - Clay Mathematics Institute claymath.org