Loading…
On 8 September 2026 OpenAI announced that a swarm of its AI agents had produced a claimed result on the Navier-Stokes existence and smoothness problem, one of the seven Clay Millennium Prize Problems.
The claim is not a proof that smooth solutions always exist; it asserts the opposite branch - finite-time blow-up, in which the computed speed of a three-dimensional fluid grows without limit in finite time, corresponding to Fefferman's alternatives (C) and (D).
The run used roughly 10,000 parallel agents over about 88 hours, exchanging 2.7 million messages and some 130 billion output tokens, producing a 166-page paper formalised in the proof assistant Lean.
Hours before the announcement, mathematician Tristan Buckmaster of NYU's Courant Institute said he and Levent Alpoge had spent over a year on closely related Euler-equation results using private AI sessions, and questioned whether their unpublished drafts had influenced the model; he stated explicitly that he was not accusing anyone.
OpenAI said no person or system searched user data to solve the problem, but later conceded it could not rule out that de-identified data from product usage had improved its models; the Clay Institute requires peer-reviewed publication and has acknowledged no solution.
A solution blows up when a quantity it describes - here the fluid velocity - becomes infinite after a finite time. The Millennium Problem asks whether smooth three-dimensional solutions always persist; proving blow-up would answer it in the negative.
Simple Analogy: A stock price that does not merely rise but hits infinity by Friday.
The Clay Mathematics Institute names Navier-Stokes existence and smoothness one of seven Millennium Prize Problems, each carrying $1 million.
Tristan Buckmaster and Levent Alpoge obtain smooth forcing results for the Euler equations, verified in Lean about a week later.
OpenAI launches its multi-agent run.
The agents reach the claimed Navier-Stokes result, roughly 88 hours into the run.
About 12 hours before the announcement, Buckmaster publicly raises the question of whether the company's model benefited from his unpublished work entered into private AI sessions.
OpenAI announces the claimed result, with a 166-page paper and a Lean formalisation.
Private American foundation that in 2000 designated seven Millennium Prize Problems, each carrying a $1 million award, and which requires peer-reviewed publication before any prize can be considered. Only the Poincare conjecture has been resolved, by Grigori Perelman, who declined the prize in 2010.
American artificial intelligence company that ran the roughly 10,000-agent system and published the claimed result; its Chief Research Officer Mark Chen said no person or AI system searched user data to solve the problem.
Institution of Tristan Buckmaster, the mathematician who raised the priority and data questions; he has worked for years on blow-up phenomena in fluid equations.
GS Paper 3 > Science and Technology > Developments in artificial intelligence and their applications; GS Paper 4 > Ethics in data use and intellectual property
General Awareness > Science and current affairs
General Awareness > Science and technology
Nonlinear partial differential equations describing the motion of viscous fluids, treating the fluid as a continuous medium.
Seven unsolved mathematical problems named by the Clay Mathematics Institute in 2000, each carrying a $1 million award.
A proof assistant in which a mathematical argument is written in machine-checkable form, so every logical step is verified automatically.
An arrangement in which many AI agents work in parallel on parts of a problem and exchange results, rather than a single model reasoning alone.