OpenAI Says 10,000 AI Agents Solved Navier-Stokes, Sparking Credit Dispute With Tristan Buckmaster

OpenAI announced that an “internal OpenAI system” had found a solution to the Navier-Stokes existence and smoothness problem, using about 10,000 AI agents that worked largely autonomously for 88 hours.

“On Tuesday, OpenAI, the developer of ChatGPT, announced in a blog post that an “internal OpenAI system” had just found a solution”

Smithsonian Magazine

The company said it verified its agents’ work with Lean, and it framed the result as a “blowup” scenario involving a vortex of fluid that spirals inward and becomes stretched out, like spaghetti.

Alerta Noticias Temuco.cl

OpenAI also said the work began on September 1 after it heard rumors that two Millennium Prize problems might have been solved, and it later realized the rumors regarded Anthropic researcher Levent Alpöge and New York University mathematician Tristan Buckmaster.

In the same reporting, OpenAI’s internal effort was described as spending “around 10,000” agents for “88 hours,” with computational power “likely cost millions of dollars,” while the Clay Mathematics Institute’s Navier-Stokes question carries a $1 million reward apiece.

The announcement triggered a dispute over whether OpenAI used Buckmaster and Alpöge’s private work, with Buckmaster questioning whether the company “could have benefited from private research conducted by him together with Anthropic researcher Levent Alpöge.”

OpenAI denied direct access, saying the human researchers and AI agents “did not see any part of their work until they published it publicly,” and it also stated that “no person or AI system searched user data to solve the problem.”

Brief IA

In a separate account of the controversy, OpenAI wrote, “While unlikely, we cannot rule out that de-identified data derived from their usage of our products helped improve our models,” while the Clay Mathematics Institute marked the question as “active” instead of “solved.”

The dispute also centered on attribution and timing, with Buckmaster describing a tense exchange in which he asked whether the model had been trained on or had access to Codex sessions, and OpenAI’s response being that “the model did not look up user data.”

The controversy raised questions about how scientific priority and credit should be handled when AI systems compete in high-stakes mathematics, with the Smithsonian noting that the achievement became contentious due to a possible association with the work of two human researchers.

“the achievement has become contentious because of a possible association with the work of two human researchers”

Smithsonian Magazine

OpenAI’s own framing emphasized that it would not seek the Clay Institute prize, while the broader debate in the reporting tied the dispute to transparency, authorship, ownership, and ethics in machine-assisted science.

The Cool Down also described unresolved questions about whether the proof will hold up under peer review and what standards should be followed when proprietary data is used in interactions that may improve other models.

Meanwhile, the reporting on the dispute highlighted that the Navier-Stokes equations are used in fluid dynamics applications such as designing aircraft and creating climate models, meaning the outcome could affect how quickly and confidently researchers rely on the underlying theory.