In fluid dynamics, there's a question that has defied mathematicians for two centuries: do the Navier-Stokes equations, which describe the motion of liquids and gases, always admit smooth, global solutions? It's one of the most famous open problems in mathematics — listed among the Clay Mathematics Institute's Millennium Prize Problems with a $1 million bounty, alongside the Riemann Hypothesis and the Poincaré Conjecture. Now OpenAI claims its AI has done it.
On September 8, OpenAI released a 166-page mathematical proof claiming that its unreleased model — via collaboration among roughly 10,000 AI agents — produced a proof of the "Navier-Stokes existence and smoothness problem" in 88 hours, complete with Lean formal verification, a rigorous framework where a computer checks every step of reasoning. If true, this would be the first time AI cracked a Millennium Prize-level problem.
The mathematical community's response, from the moment of release, has been cautious. The Clay Mathematics Institute still lists the problem as unsolved and stresses that the official verification process has not begun. The controversy centers on three points. First, credit attribution: does an AI-generated proof count as "solved by humans"? This touches on award rules — the institute has historically only recognized formal achievements by human mathematicians. Second, while the 166-page proof comes with Lean verification, that verification targets the AI's internal proof code; the mathematical community hasn't independently reviewed the logic chain. Third, OpenAI has a contested track record on mathematical claims — its models have previously been accused of "gaming" benchmarks rather than truly understanding mathematics.
OpenAI, for its part, is highly bullish. President Greg Brockman called this "an important step forward in AI scientific reasoning," linking it to the company's research on "recursive self-improvement" — if AI can prove theorems, it can verify its own reasoning, moving toward stronger autonomy. In internal demo videos, the model showed multi-step reasoning on nonlinear partial differential equations that, superficially, resembles how human mathematicians work.
For ordinary readers, what matters isn't "who gets the million dollars" but a larger trend: AI is moving from being a "tool for solving equations" to a "discoverer of mathematics." Over the past year, from Fields Medalist Terence Tao publicly using AI-assisted research to labs adopting math as a benchmark for reasoning ability, the boundary between AI and mathematics is rapidly blurring. The Navier-Stokes equations carry enormous engineering significance — weather forecasting, aircraft design, and ocean current simulation all depend on them — and if genuinely proven, AI's contribution would be undeniable.
Still, as many mathematicians remind us: between an AI-generated proof and a human-endorsed one lies not computing power, but verifiability and comprehensibility. Whether the million-dollar check gets written depends on the institute's independent review — a process that could take far longer than 88 hours.