OpenAI has announced that an internal AI system produced a proof for the Navier-Stokes existence and smoothness problem, one of seven Millennium Prize Problems set by the Clay Mathematics Institute in 2000. The announcement has been met with excitement over what could be a historic mathematical breakthrough, alongside pointed questions about whether the AI system drew on unpublished human research to get there.
What OpenAI Claims to Have Solved
The Navier-Stokes equations, dating back to nineteenth-century work by Claude-Louis Navier and George Gabriel Stokes, describe the motion of fluids and have remained a central unresolved question in mathematics for roughly 90 years. OpenAI says its system produced both an analytical proof and a formal verification in the programming language Lean, showing that a smooth, three-dimensional fluid at rest, under a smooth applied force and with finite energy throughout, can develop a singularity in finite time.
The company said the proof came from an internal model more advanced than its current publicly available system. According to OpenAI, groups of autonomous AI agents, reportedly numbering in the thousands, worked over roughly 88 hours to arrive at the result, following an earlier effort in which around 100 agents spent about 50 hours producing a related disproof involving the Euler equations.
Why the Timing Raised Questions
OpenAI has said its push on Navier-Stokes began on September 1, after the company heard rumors that two Millennium Prize problems had been solved elsewhere. Those rumours reportedly traced back to independent work by mathematicians Theodore Buckmaster and Levent Alpöge, who had been using AI tools, including models from a rival lab, to work through a simpler version of the same problem. Their approach reportedly built on earlier methods developed by other researchers in the field.
The overlap between OpenAI’s internal timeline and the unpublished work of outside mathematicians has raised a central concern: whether an AI lab’s own systems could have had indirect access to research that had not yet been made public. OpenAI has stated that its agents worked only from cached internet data and code execution tools, did not view the outside mathematicians’ work through any channel before it was released publicly, and did not access specific user data in the course of solving the problem. The company also said it offered the two mathematicians a concurrent release of results, along with visibility into the prompts used and an early look at the proof.
A Trust Problem for AI-Assisted Science
The episode highlights a broader tension now facing research fields that increasingly rely on AI tools: researchers working on unpublished, high-value problems may have no way of confirming whether the same lab whose tools they are using is also training or running systems that could arrive at similar results independently, or through some form of indirect exposure to their work.
Some mathematicians in the field have voiced discomfort with what they describe as a shift toward frenetic competition, warning that the unchecked use of AI systems to chase priority on famous problems could reduce parts of mathematics to a race for output rather than deliberate, verifiable progress. Others have called the result a genuinely remarkable achievement, regardless of the surrounding controversy, and have noted that the Clay Mathematics Institute itself has acknowledged the significance of the moment.
What Happens Next
OpenAI has published its write-up and the formal Lean proof publicly, allowing independent mathematicians to verify the result. Formal verification through Lean gives the mathematics community a higher degree of confidence that the underlying logic holds, even as questions about the research process behind it continue.
For African and Middle Eastern tech and research communities increasingly exploring AI-assisted research and academic tools, the controversy is a preview of governance questions that will likely follow as AI systems get pulled deeper into scientific and mathematical discovery: how research provenance is protected, how credit is assigned, and how much visibility outside researchers should have into what AI labs’ internal systems are trained on or exposed to.
