Hype check: “AI solved the Navier–Stokes Millennium Prize problem”
OpenAI's proof needs a designed outside force pushing the fluid, which the prize rules allow. A fluid left alone is still an open question.
The claim in full “AI solved the Navier–Stokes Millennium Prize problem (OpenAI, September 2026).”
What the data shows
The claim: on 8 September 2026, OpenAI announced that an AI system had proved the 3D Navier–Stokes equations (which describe how fluids such as water and air move) can break down in finite time. Many reports said AI had solved a Millennium Prize problem.
What’s true. The proof is real, and the official rules allow its approach. The problem statement, written by Charles Fefferman in 2000, offers four ways to win. Two ask for proof that a fluid with no outside force stays smooth forever. The other two ask for a breakdown, and they allow a smooth outside force. OpenAI used that option: its fluid starts at rest and is pushed by a carefully designed force. Diego Córdoba argued that every real fluid is under some force, so including one is natural.
What experts say is still open. Scientific American reported that most researchers had been looking for a breakdown that comes from the fluid’s own motion. Luis Silvestre (University of Chicago) put it bluntly: “The most important problem is unsolved”. Two follow-up papers mark the limits. Constantin, Ignatova and Vicol showed that constructions sharing two key features of OpenAI’s need a force that is neither zero nor real-analytic (a very smooth kind of function) near the breakdown point. That rules out a direct extension to the force-free problem. Bastounis, Circelli and Hansen argue that OpenAI’s machine-checked file (written in the Lean proof language) and its paper make different claims: the check guarantees the Lean statement, not that it says what the paper says.
What it means
An AI system has produced a real, machine-checked result on a famous fluid problem, using an option the prize rules allow. It has not answered the question most mathematicians care about, whether a fluid left alone can break down, and no prize has been awarded. Headlines saying AI “solved Navier–Stokes” overstate it.
Caveats
- The Clay problem statement genuinely allows forcing, so calling it a loophole is a value judgment. Clay has said its evaluation will be deliberately slow, and no prize has been awarded.
- Kevin Buzzard and others hold that the Lean-checked headline statement is the right one for this problem. The faithfulness critique concerns the paper’s intermediate claims, and its authors explicitly do not claim the proof is wrong.
- OpenAI also says its agents found blow-up for the 3D Euler equations with no force as an intermediate step (per ScienceABC); that claim has not been independently checked.