Anthropic AI Claude Claims to Solve Math’s ‘White Whale’

Anthropic AI Claude Claims to Solve Math's 'White Whale'

I was halfway through my coffee when a link pinged across my feed and the math world stilled. For a few minutes I felt the room shrink: one man, one Lean file, one claim that might topple a century-and-a-half-old mountain. I want you to sit with that for a second, because whatever you feel — hope, skepticism, alarm — you’re not alone, and this matters.

On a Sunday night, Levent Alpöge says he finished the proof.

I’ve read the note he posted, and I’ve followed the Lean repository. Alpöge, a Turkish-American mathematician working at Anthropic, says he used Claude to help build a proof of the Riemann hypothesis — the problem that has shadowed prime numbers since 1859. If verified, this would be one of those rare scientific moments where a single post reorders how we think about human and machine labor in pure research.

Did AI really solve the Riemann hypothesis?

Short answer: not yet. Long answer: Alpöge published a formalization in Lean and credited Claude for substantial assistance; peers and the Clay Mathematics Institute still need to vet the work. You should expect months, not hours, of verification — Lean files are rigorous, but the community must trace every inference and dependency before anyone calls this settled.

At a company desk, the timing was awkward and obvious.

Alpöge says he texted Anthropic CEO Dario Amodei the night he reached the endpoint. Anthropic just filed a record-breaking IPO, and any major claim linked to the company will be read through a market lens. He acknowledged the timing was “horrible,” and people will interpret that in light of incentive and optics — which is part of why independent confirmation matters more than glow or press releases.

In lecture halls and Slack channels, the old rivalry flared up.

You probably remember OpenAI’s Navier–Stokes announcement this fall. That episode left many mathematicians bruised: OpenAI said it solved a Clay Millennium Problem with a swarm of agents, then faced accusations that prior academic work wasn’t credited. Now another bold claim — this time linked to Anthropic — has reopened fault lines about credit, collaboration, and corporate influence in fundamental science.

Subscribe to our daily AI newsletter
The 5-minute AI newsletter. The stories that matter, in your inbox every weekday morning.
Subscribe Now

How did Claude help produce the proof?

Anthropic’s Claude is a large language model designed for reasoning and code generation. Alpöge says he used Claude to explore ideas, write machine-checkable arguments, and accelerate the Lean formalization. That sounds like collaboration; whether Claude suggested the crucial insight or simply executed it is exactly the kind of question peer reviewers will agonize over.

In the small print of academic culture, credit still matters.

Tristan Buckmaster and others raised concerns about OpenAI’s Navier–Stokes claim, saying prior academic work was used without proper attribution. I’ve spoken to mathematicians who worry AI-enabled projects will blur lines between inspiration, assistance, and appropriation. You and I can both see the stakes: intellectual ownership, career incentives, and public trust hinge on how the next weeks play out.

At a deeper technical level, Lean matters more than soundbites.

Lean is a proof assistant that forces arguments into a machine-checkable form. A verified proof in Lean is not a press release; it’s a dense package of definitions and lemmas that others can inspect and replay. If the community can load Alpöge’s files and reproduce every step, that gives the claim genuine weight — but reproducing and accepting are not the same thing.

What does the Clay Mathematics Institute need to declare a solution?

The Clay Institute requires a published proof and a period for independent verification before awarding the $1 million (€920,000) prize. That process is deliberate and strict, because a single mistake would ripple across number theory. Expect referees to comb the Lean formalization and the surrounding arguments — and to argue intensely about which parts are novel.

I’ve tried to keep my skepticism and my curiosity balanced because you deserve a clear read: this could be a turning point for mathematics, or it could be a high-profile episode that teaches the field how to scrutinize AI-assisted claims. Claude helped, Alpöge asserts; Anthropic stands to gain; the community will decide — but tell me, when a machine helps chase the white whale of mathematics, who should get the laurels and who should carry the bill for errors?