On Monday, OpenAI posted a page on its website claiming that one of its reasoning models had produced a proof of the Navier-Stokes existence and smoothness problem — one of the seven Millennium Prize Problems, with a $1 million bounty attached. The proof, the company said, was being prepared for expert review. It did not post the proof itself.

That last sentence is the whole story.

The Clay Mathematics Institute, which administers the prize, still lists Navier-Stokes as unsolved. Its rules require publication in a refereed journal and a two-year waiting period before any award. OpenAI’s announcement satisfies neither condition. What it satisfies is something else entirely.

The Verification Gap

Mathematics runs on a simple norm: a proof is a public document. You don’t get to claim a theorem and keep the argument to yourself. The entire edifice — from Euclid to Perelman — rests on the idea that anyone, anywhere, can check the work. That’s what makes mathematics different from a pharmaceutical trial or a proprietary algorithm. The proof is the product, and the product is public.

OpenAI has now proposed a different model: the proof is the product, and the product is private.

This is not a small procedural quibble. The Clay Institute’s two-year waiting period exists precisely because verification takes time and because the community, not the claimant, decides what counts as a proof. When Grigori Perelman solved the Poincaré conjecture, he posted his papers to arXiv and let the world tear them apart for three years before the prize was awarded. He didn’t hold a press conference.

OpenAI’s announcement inverts that order. The claim comes first; the verification, if it ever comes, will follow. And in the meantime, the claim itself does work — it moves markets, it shapes narratives, it positions the company as the entity that solved one of the hardest problems in mathematics.

The Codex Question

Then there’s the other wrinkle. A mathematician who had been working on the Navier-Stokes problem has publicly wondered whether OpenAI’s model had access to his notes — notes he had stored in Codex, OpenAI’s own code repository tool. The company has not addressed the question directly.

This is the part that should make everyone uncomfortable, regardless of what you think about AI. If a company trains a model on the world’s mathematical literature, and a researcher’s unpublished notes happen to be in that corpus, and the model then produces a proof that resembles those notes — whose proof is it?

The $1 million prize is almost beside the point. OpenAI’s market capitalization moves by billions on a good day. What’s at stake is attribution, priority, and the basic question of whether a proof produced by a system trained on other people’s unpublished work can be claimed as a novel result.

One researcher who was asked to sign a non-disclosure agreement before reviewing the proof put it this way: “I’ve reviewed a lot of mathematics in my career. I’ve never once been asked to sign an NDA to look at a proof. That’s not how this works. That’s how you protect a trade secret.”

The Real Prize

Here’s what the conventional take misses. The story is not “AI solved a Millennium Prize problem.” The story is that a company has decided it can claim a mathematical result without subjecting it to the one process that makes mathematical results meaningful.

The $1 million is a rounding error. The real prize is the headline — the ability to say, in a funding round or an earnings call or a product launch, that your model solved Navier-Stokes. Whether the proof holds up is a question for later, and by the time it’s answered, the narrative will have already done its work.

This is the financialization of mathematical truth. A proof used to be a public good — something that, once published, belonged to everyone. OpenAI is treating it as a private asset, to be revealed on the company’s schedule, under the company’s terms, with the company’s name attached.

The mathematicians will eventually sort out whether the proof is correct. That’s what they do, and they’re good at it. But the deeper question — whether the institution of mathematics can survive a company that treats verification as optional and attribution as negotiable — is not one the Clay Institute can answer with a two-year waiting period.

It’s a question about what happens when the people who make claims stop caring whether anyone can check them.

Sources