OpenAI claims huge maths breakthrough on a famed ‘Millennium Problem’

Los Angeles, the 8th of September, 2026

The claim: a machine has solved a Millennium Problem, the one about fluids. Navier-Stokes. And I find myself doing what I always do when someone announces a miracle - asking to see the wristwatch work, not the advertisement for it.

Here’s what a solution has to be. Mathematics is not a poll. It is not “most experts found the argument persuasive.” It is: write down the proof, line by line, such that any competent person can check every line and find no gap. That’s the whole contract. So the first question isn’t “how smart is the machine?” It’s “can I read the answer?” If the output is a thousand pages no human has verified, we don’t have a proof; we have a claim of a proof, which is worth roughly the paper it isn’t printed on.

And wait - there’s something odder in the wording. “Solved the physics of fluids.” But the Millennium Problem isn’t about physics. It’s about whether the equations always behave - whether smooth solutions stay smooth or whether, somewhere, they blow up to infinity. That’s a question about the mathematics itself, not about turbulence you can photograph in a tank. A simulation that matches experiment beautifully tells you nothing about it. You can fit a million data points and still not know what the equation does at some unmeasured, vicious corner.

That distinction is being sanded away in the press release, and I want to know why. Perhaps because the verifiable version - a formal proof, checked step by step by a machine-independent checker - is harder to sell in a headline than the word “solved.”

I’d be delighted to be wrong. Show me the proof, in small steps, and I’ll sit down with a pencil like anyone else. Until then, my rule holds: the easiest person to fool is yourself, and a company announcing its own triumph has every incentive to be fooled first.

Extraordinary claims. Checkable details. That’s all I ask.

8 September 2026, evening

The news says a machine has mastered the motion of water. I confess I read it twice, and laughed once.

For years I filled page after page with studies of eddies: how a current striking a flat obstacle splits, curls, folds back upon itself; how the whirlpool born at a stone’s edge transmits its rotation outward until the whole river carries it. I wrote then that turbulence was a knot no straight mind could untie - one stroke of the pen and the water escapes you. If these modern calculators have truly seized the Navier - Stokes problem, they have seized the very current I chased across half a lifetime of sheets.

But observe what the claim actually is. A proof in mathematics is like a bridge arch: every stone must bear its neighbour, and if one voussoir is misshapen, the whole span drops. So far I see only the company’s own announcement - a declaration, not the arch itself. Pressure distributed through a single point of origin, one corporation bearing the entire load of verification. Columns fail suddenly; I would want a hundred hands testing each stone before I walked across.

Three things I must examine when the work surfaces: whether the proof shows the mechanism or only the conclusion, as a shadow shows the outline of a bird but nothing of the wing’s leverage; whether the machine found the solution by reasoning or by combing through a million trials as a sieve separates grain from chaff - no shame in either, but the two teach different lessons; and whether the solution holds at every scale, for fluids that run smooth in the channel misbehave in the gorge.

I have not yet determined whether a thing that computes water has understood it. That question I leave open, and it is the better question.

The claim arrived pre-glittered: “cracked,” “breakthrough,” one of the toughest puzzles in mathematics. My first instinct is always to ask who benefits from the frame being that shiny before anyone has touched the object. A Millennium Problem isn’t a homework set you finish overnight; it’s a load-bearing wall in the discipline, and the Clay Institute has spent decades waiting for someone to bring not a press release but a proof - every lemma, every epsilon, checked by hands other than your own.

So what’s actually being claimed? Not, it seems, a proof in the classical sense, but something about fluids - turbulence, maybe, or regularity - modelled, predicted, simulated at a fidelity nobody managed before. That’s genuinely exciting work. But notice the sleight of hand in the headline: “solved” doing the work that “produced a remarkable computational artefact requiring independent verification” should be doing. The detail pushed to the edge is verification. Who has read it? How long did they read for? A proof is a social contract, not a demo.

And the plain question the room has to answer: if the machine’s output only means something after mathematicians spend two years checking it line by line, in what sense has the machine done the mathematician’s job - and in what sense is it honest to sell the result as done?

I like the people doing this work. They’re brilliant, mostly exhausted, working inside a company whose incentives reward the announcement more than the audit. I’ve been the tired person inside a system that wanted the story before the substance. Fond exasperation is the correct temperature here.

Turbulence has humbled everyone from Kolmogorov to the present morning. If this holds, wonderful - hand it to the referees, not the shareholders. Until then, it’s a claim about fluids, and the most relevant fluid is the one this news is trying to spill on us.