OpenAI’s Navier-Stokes proof is a company-announced mathematical result, not yet an independently confirmed resolution of a Millennium Prize Problem. OpenAI said September 8 that an internal AI system produced an analytical proof, plus a Lean formalization, showing a particular three-dimensional fluid flow can develop a finite-time singularity. The announcement immediately met questions about the work’s provenance after two researchers published a related result the previous day.
The distinction is doing a lot of work here. OpenAI has released its writeup and formalization, but the material available does not establish that mathematicians outside the company have validated the claimed proof. It also does not settle whether product-usage data affected the underlying model or whether other researchers’ work influenced the result.
What does OpenAI claim its Navier-Stokes proof shows?
According to OpenAI, its proof establishes that an initially smooth fluid at rest, subject to a smooth applied force, can form a singularity in finite time while retaining finite energy. In the company’s description, a singularity means fluid speed rises without bound within a finite period, despite viscosity, which normally smooths motion.
The Navier-Stokes equations describe the movement of fluids, including liquids and gases. Their existence-and-smoothness question has remained unresolved for roughly 90 years, OpenAI said. The Clay Mathematics Institute listed it among seven Millennium Prize Problems in 2000, each carrying a $1 million award.
OpenAI said its result corresponds to statements C and D in the problem’s official formulation. It also said it will not seek the $1 million prize.
OpenAI’s account of how the result was produced
OpenAI said it began training a new internal model on August 28 and described it as more capable than GPT-6 Astra. After hearing rumors on September 1 that two Millennium Prize Problems had been resolved, the company said it launched groups of agents against open prize problems and several other difficult questions.
The Navier-Stokes effort used roughly 10,000 concurrent agents, according to OpenAI. The company said those agents reached their claimed result about 88 hours after the first ones were launched; Lean formalization and verification then took another 17 hours using GPT-6 Astra. That is OpenAI’s account of its process, rather than independent verification of the mathematics.
Why are researchers questioning the announcement?
New York University mathematician Tristan Buckmaster and Anthropic researcher Levent Alpöge released findings on a related problem one day before OpenAI’s announcement, The Verge reported. Buckmaster said OpenAI’s Navier-Stokes proof followed a route that he and Alpöge had been investigating with OpenAI’s Codex and Anthropic’s Claude.
Buckmaster said he asked OpenAI whether its model had been trained on, or had access to, the pair’s Codex sessions, where they had put drafts of their work. He said OpenAI told him the model did not look up user data, but that he did not receive an answer when he asked specifically about training. That is an allegation and an unanswered question, not evidence that OpenAI accessed the sessions.
OpenAI said no specific user data was accessed to solve the problem. The company also said it could not rule out, though it considered it unlikely, that de-identified data derived from product use had helped improve its models. Sébastien Bubeck, an OpenAI technical staff member, said the company did not see Buckmaster and Alpöge’s work until its public release and that the proofs and the precise results are materially different.
For now, the announcement contains three separate questions: whether the proof holds up, whether any de-identified usage data mattered, and whether the related work had any influence. The published accounts do not answer any of them conclusively.
This story draws on original reporting from The Verge.
