OpenAI claims proof of fluid equation blowup as mathematicians allege credit theft
A computer-verified proof of finite-time singularities in the Navier-Stokes equations triggers a research priority dispute and software training-data scrutiny.
Original cover art, generated for this story. THE VISSION does not republish third-party press imagery.
- OpenAI announced that an internal AI swarm of 10,000 agents running for 88 hours had constructed a computer-verified Lean proof showing finite-time blowup for the 3D incompressible Navier-Stokes equations.
- NYU mathematician Tristan Buckmaster and Anthropic researcher Levent Alpöge allege that OpenAI accelerated its project after hearing rumors of their own imminent breakthrough.
- Buckmaster raised privacy concerns over their work-in-progress drafts stored in OpenAI's Codex model, questioning if their private data was harvested to train OpenAI's systems.
OpenAI has claimed a monumental mathematical breakthrough by announcing a complete, computer-verified solution to the Navier-Stokes existence and smoothness problem, one of the seven million-dollar Millennium Prize Problems. The company stated that an internal, unreleased AI system, utilizing a highly advanced swarm of approximately 10,000 autonomous agents, successfully constructed a mathematical proof demonstrating that the three-dimensional incompressible Navier-Stokes equations can develop a singularity—or 'blowup'—in finite time. The entire computational process reportedly took 88 hours, followed by an additional 17 hours for the system to translate and verify the mathematical logic within the Lean interactive theorem prover.
The announcement has immediately sparked an intense controversy within the international mathematics community, turning what would be a landmark scientific milestone into a bitter dispute over academic priority and ethical data collection. Tristan Buckmaster, a mathematics professor at New York University, and Levent Alpöge, an AI researcher and mathematician at Anthropic, had been working on a related breakthrough for nearly a year. On the same day as OpenAI's announcement, the pair published three papers proving finite-time blowup for the 3D incompressible Euler equations with smooth forcing—a key mathematical stepping stone toward Navier-Stokes. Buckmaster has publicly alleged that OpenAI aggressively accelerated its Navier-Stokes project only after learning rumors of their impending results.
Crucially, the dispute has highlighted deeper concerns about data privacy and the provenance of high-quality training material used by commercial AI laboratories. Buckmaster and Alpöge had been utilizing OpenAI’s Codex model to store and refine their work-in-progress drafts. Buckmaster raised concerns that OpenAI may have scanned their private Codex files to inform and jumpstart its own proof. While OpenAI researcher Sébastien Bubeck denied any data theft or improper access, the company's official blog conceded that rumors of a breakthrough inspired their sprint, and they could not rule out that the researchers' usage of OpenAI products helped improve the underlying models. Fields Medalist Terence Tao described the AI-generated proof as impressive but urged caution until the 100-page document undergoes rigorous human peer review.
If verified, this proof represents the second solution to a Millennium Prize Problem and a fundamental shift in fluid dynamics, suggesting fluid velocities can theoretically reach infinity under extreme conditions. However, the controversy exposes a critical vulnerability in the relationship between researchers and AI labs. If scientists cannot trust commercial AI models to keep their work-in-progress private, the resulting chilling effect could stall high-stakes scientific collaboration and accelerate a crisis of trust in AI-driven scientific discovery.
Will the Clay Mathematics Institute formally award the $1 million prize to OpenAI, and how will human peer reviewers evaluate the 100-page AI-generated proof?
Still open. When the paper finds out, it will say so here and on the open questions page — including if it got this wrong.