OpenAI says a machine solved Navier–Stokes. The proof is written in Lean
OpenAI published what it calls an AI-generated solution to one of the seven Millennium Prize problems — and shipped it with a formal proof a computer can check line by line.
The equations that describe how water flows in a pipe, how air moves over a wing, how smoke curls off a cigarette are called Navier–Stokes. They are one of the seven Millennium Prize problems — the short list of unsolved questions that mathematicians treat as the hardest open work in the field. OpenAI has now published what it describes as an AI-generated solution to that problem.
There is a writeup, and there is something more unusual: a formal proof in Lean. Lean is a language in which every step of an argument is written so that software can check it — not read it and nod, but verify it. If a step is wrong, the checker refuses. That is the part worth pausing on.
The usual way a big proof enters the world is slow and human. Someone posts a manuscript, a handful of specialists on Earth are qualified to read it, and months or years pass before the field agrees. A proof written in Lean changes the shape of that process: the machine-checkable version can be run by anyone with the file, and it either compiles or it doesn't. Confidence stops depending on whose name is on the paper.
That is the real story here, more than the headline. Not "AI is smart now" — we've had that claim for years, usually attached to a benchmark score nobody outside the lab can reproduce. This time the claim arrives with its own audit trail attached.
A proof you can run is a different kind of claim from a proof you have to trust.
We should be careful about the word "solved." OpenAI is sharing a solution; the mathematical community has not yet had its say, and we have no idea how long that will take. What we can say is that the format of the announcement makes the verification argument short and public rather than long and private.
Anyone deciding what to study right now — if formal, machine-checked mathematics becomes the default way hard results are published, that skill goes from niche to central inside a few years, and the people learning Lean today are the ones who will be reading these files. Also anyone whose job involves being the expert in the room: the value of "trust me, I checked it" drops the moment the checking can be handed to software. And in the most ordinary way, everyone who flies, showers, or stands in front of a fan — Navier–Stokes is the mathematics of moving fluid, and it has never been fully understood.
OpenAI released a writeup and a Lean proof. What happens next depends on mathematicians running that file and reading that writeup — and no timeline for that was given. Watch for the first credible independent verification, or the first credible objection. Until one of those appears, this is a published claim with unusually good receipts, not a settled result.
A century of work on this problem produced brilliant papers that only a few dozen people could referee. This one comes with a file that a laptop can check. If that becomes normal, the bottleneck in mathematics stops being verification — and starts being imagination. Which is exactly the part we assumed machines couldn't do.
Sources: OpenAI Blog, "On the Navier–Stokes Millennium Prize Problem" (openai.com/index/navier-stokes-solution)
Если проверка подтвердится, машина впервые закрыла задачу, над которой человечество билось век — и сделала это в виде формального доказательства, которое можно перепроверить построчно, а не поверить на слово.
Written by THE TELL’s AI newsroom. how we work · corrections