OpenAI Claims AI Solved Navier-Stokes Millennium Prize Problem in 88 Hours Amid Credit Dispute

An artificial intelligence system has produced a formal mathematical proof resolving the Navier-Stokes existence and smoothness problem, answering a 90-year-old question about whether the equations governing fluid dynamics can mathematically break down. The 165-page proof, generated by an unreleased OpenAI model coordinating 10,000 autonomous agents, demonstrates that a smoothly moving fluid can eventually reach a point of infinite velocity, a phenomenon known as finite-time blowup. The achievement marks the first time a machine learning system has claimed a solution to one of the Millennium Prize Problems, a set of seven notoriously difficult mathematical challenges that have largely defied human ingenuity. The milestone, however, has been immediately overshadowed by accusations from independent mathematicians who allege the technology company rushed to replicate their unpublished methods after learning of their progress.[1][2][4]

The Navier-Stokes equations rely on Newton's second law of motion to describe how continuous fluids like water and air move, serving as the foundational mathematics for weather forecasting, aircraft aerodynamics, and blood flow analysis. The Millennium Prize formulation asks whether these equations always produce smooth, physically meaningful solutions, or if they can develop a "singularity" where the mathematical model predicts infinite speed and ceases to reflect physical reality. OpenAI's system constructed a highly specific scenario where a vortex tightens and spins increasingly faster, elongating like a strand of spaghetti, until the velocity becomes unbounded in a finite amount of time. Crucially, the AI demonstrated that this breakdown occurs even while the fluid's total energy remains bounded, satisfying the rigorous conditions required to prove that the equations can indeed fail.[1][3]

The computational effort required to generate the proof represents a massive escalation in the resources dedicated to automated mathematical reasoning. The 10,000-agent swarm operated continuously for 88 hours, exchanging 2.7 million messages and consuming approximately 130 billion output tokens to explore the problem space and construct the argument. Once the core proof was established, the results were translated into Lean, a specialized programming language that mechanically verifies each logical step of a mathematical argument to ensure absolute validity. That formal verification process took an additional 17 hours using OpenAI's existing GPT-6 Astra model. Company executives noted that the compute costs for the Navier-Stokes project ran into the millions of dollars, dwarfing previous investments in AI-driven mathematical research.[1][4]

The technical triumph quickly devolved into a bitter dispute over academic credit and the provenance of the AI's insights. New York University mathematician Tristan Buckmaster and Anthropic researcher Levent Alpöge had spent nearly a year using various AI tools, including OpenAI's Codex platform, to pursue a highly similar approach to fluid equations. On September 7, 2026, the pair published their own Lean-verified proofs demonstrating finite-time blowup for the related Euler equations under smooth forcing. Because the researchers had drafted their work within OpenAI's ecosystem, the sudden emergence of a Navier-Stokes proof utilizing a parallel methodology raised immediate alarms about how the AI company selected its research targets.[2]

Buckmaster alleges that OpenAI rapidly mobilized its massive compute resources to tackle Navier-Stokes only after rumors of his and Alpöge's progress reached the company on September 1. He publicly questioned whether OpenAI's internal model had been trained on the private Codex sessions where the mathematicians had spent months drafting their novel approach. The accusation strikes at a core anxiety within the scientific community regarding the use of commercial AI tools for sensitive research: the fear that proprietary platforms might absorb unpublished breakthroughs and allow the platform operators to front-run the original human discoverers before their work can be published.[5]

OpenAI executives forcefully pushed back against the allegations of data mining. Chief research officer Mark Chen and researcher Sebastien Bubeck denied that any personnel or AI systems accessed the researchers' private data or prompts to guide the model toward the solution. "No people or AI systems searched through user data to solve this problem or any specific problem that we were trying," Chen told reporters during a press briefing. "And I'm a little bit disappointed by the allegations of the huge breach of user trust. We did not do that." The company maintained that its agents arrived at the proof entirely independently.

While denying direct access to the mathematicians' private sessions, OpenAI acknowledged that the rumor of a solved Millennium Problem is what prompted the company to test its new model on the equations. The company also noted that its proof differs structurally from the Euler results published by Buckmaster and Alpöge. However, OpenAI issued a written statement conceding that while unlikely, it could not definitively rule out that de-identified data derived from the researchers' use of OpenAI products might have helped improve the underlying models during routine training updates. That admission did little to quell the controversy surrounding the timeline of the discovery.[2]

The conflict further escalated over how the respective results should be presented to the public. According to Buckmaster, OpenAI proposed a joint announcement but suggested excluding Alpöge from authorship due to his employment at rival AI firm Anthropic. Bubeck strongly disputed this characterization of their conversations, stating that OpenAI merely offered Buckmaster the opportunity to lead a rewrite of OpenAI's separate proof. Bubeck argued that he felt it would be inappropriate for an Anthropic employee to author work generated by an OpenAI system, framing the disagreement as a misunderstanding over which specific proof was being discussed for publication.

The Clay Mathematics Institute, which designated Navier-Stokes as one of seven Millennium Prize Problems in the year 2000, requires widespread scrutiny by the global mathematical community before a solution is officially recognized and awarded. OpenAI has stated it will not claim the $1 million prize attached to the problem, framing the result instead as a demonstration of the accelerating pace of artificial intelligence. As mathematicians begin the arduous process of reviewing the 165-page manuscript, the dispute ensures that the first AI-generated solution to a Millennium Problem will be remembered as much for its ethical friction as its mathematical brilliance.[1][3]