The Navier–Stokes equations describe how fluids move. They are used to model phenomena ranging from water flowing through pipes to air moving around aircraft and turbulence developing in the atmosphere.
The equations themselves were known more than a century ago. The open question is whether smooth, physically reasonable solutions are always smooth in three dimensions – or whether some initial conditions can lead to singularities, sometimes called mathematical “blowups.”
The problem is difficult because fluid motion can become extraordinarily complex. Small changes in initial conditions may lead to dramatically different behavior, particularly when turbulence is involved.
Why the claim matters
The Navier–Stokes question is not merely an abstract puzzle. A rigorous solution could deepen our understanding of turbulence, weather systems, ocean currents, aerodynamics and industrial fluid dynamics.
However, the practical consequences would probably not appear overnight. Engineers already use numerical approximations and simulations to model fluids. A proof would not automatically produce better aircraft, more accurate weather forecasts or new energy technology.
Its immediate importance would be mathematical: it would settle a foundational question about the equations used to describe one of nature’s most difficult systems.
OpenAI reportedly said its model used a large collection of agents that could read information, write and run code, and coordinate different lines of reasoning. The company also released a formal proof written in Lean, a programming language used to verify mathematical arguments.
That is potentially important because formal verification can expose missing assumptions, invalid logical steps and hidden gaps. At the same time, a Lean file is not automatically a correct proof. The underlying argument still needs to be understood, checked and accepted by specialists.
For readers interested in how AI systems are evolving beyond ordinary chatbots, AI Smart Core’s coverage of current AI developments provides useful context.
Why mathematicians are asking questions
The announcement came amid an unusual situation involving mathematicians Tristan Buckmaster and Levent Alpöge.
Buckmaster said that he and Alpöge had been working on the Navier–Stokes problem and had used several large language models, including OpenAI’s Codex, during their research. They were preparing to discuss their own work when they learned that OpenAI was preparing a related announcement.
Buckmaster has emphasized that he has not seen OpenAI’s proof and is not accusing the company of using unpublished research. His concern is that the timing raises questions about how AI research systems obtain information and how much human work may influence their results.
OpenAI said its work began on September 1 after the company heard a rumor that another Millennium Prize Problem might have been solved. The company also said its agents did not access the researchers’ private work and did not see their material before it was made public.
Still, OpenAI acknowledged that it could not completely rule out the possibility that de-identified data derived from product usage had helped improve its models.
That admission highlights a broader issue for AI-assisted research: proving that an AI system reached a result independently may be difficult when the system has been trained on enormous quantities of human-created material.
The difference between discovering an answer and proving it
AI systems can be extremely effective at identifying patterns, proposing conjectures and exploring large numbers of possible approaches. They can also generate code, manipulate symbols and search through mathematical structures much faster than a human working alone.
But mathematics depends on more than a plausible conclusion. A valid proof must be precise, complete and based on clearly stated assumptions.
This creates several challenges:
- Hidden assumptions: An AI-generated argument may rely on a condition that was never established.
- Invalid generalization: A pattern that works in many examples may fail in an untested case.
- Unreadable reasoning: A formal proof can be technically valid while remaining difficult for humans to interpret.
- Verification complexity: Independent mathematicians must confirm both the formal code and the mathematical ideas behind it.
- Reproducibility: Other researchers should be able to inspect the method and obtain the same result.
The strongest outcome would be a combination of machine verification and human understanding. A computer can check millions of logical steps, while mathematicians can determine whether the argument is meaningful, elegant and properly connected to the original problem.
Why Terence Tao’s criticism is significant
Fields Medalist Terence Tao compared the situation to watching the beginning and end of a movie while missing the central story.
His concern is not that AI should be excluded from mathematics. Instead, he has warned that an intense race to produce headline-making solutions could reduce mathematical research to a competition over output rather than understanding.
That criticism reflects an important tension. AI may help researchers solve problems that would otherwise take decades, but mathematics is also valued for the ideas that make a proof understandable and reusable.
A proof that no human can explain may still be formally correct. Yet it may offer less value to the wider mathematical community than a shorter argument that reveals why the result is true.
This debate is closely connected to the wider development of AI agents, which increasingly perform multi-step tasks instead of simply generating one response. AI Smart Core’s case studies on AI tools and automated workflows offer additional background on this shift.
Has OpenAI officially solved a Millennium Prize Problem?
Not yet.
In 2000 the Clay Mathematics Institute announced the seven Millennium Prize Problems and $1 million for each of them. Of the seven, only one, the Poincaré Conjecture, has been officially solved.
For a solution to qualify, it must be made publicly available, withstand professional scrutiny and receive acceptance from the mathematical community under the institute’s rules.
OpenAI has said it does not intend to claim the Millennium Prize for its Navier–Stokes result. That decision may reflect the difference between producing a promising proof and completing the lengthy process required for official recognition.
Until independent experts review the work, the most accurate description is that OpenAI has presented a potentially groundbreaking solution to the problem.
What this means for the future of AI research
The announcement points toward a future in which AI systems act less like digital assistants and more like research collaborators.
A large-scale mathematical AI system could:
- Search through thousands of possible proof strategies.
- Use code to test conjectures and counterexamples.
- Divide a difficult problem into smaller tasks.
- Ask specialized agents to check separate parts of an argument.
- Convert the final reasoning into a formally verifiable proof.
This approach could accelerate progress in mathematics, physics, chemistry and computer science.
It also raises difficult questions about authorship, credit and accountability. If thousands of AI agents contribute to a discovery, who should be considered the author? How should researchers disclose the role of training data? And what happens when an AI system produces a correct result that humans cannot easily explain?
The Navier–Stokes claim may ultimately be remembered not only for its mathematical content, but also for forcing the research community to define standards for AI-generated discoveries.
The bottom line
OpenAI’s announcement could represent a major advance in automated mathematical reasoning, but the claim still requires independent verification.
The most important development is not simply that an AI system may have found a solution. It is that AI systems are beginning to operate across long chains of reasoning, code execution and formal verification at a scale that would be difficult for a human team to match.
If the proof survives expert review, it could mark a turning point for mathematics. If it does not, the episode will still provide a valuable lesson: producing a convincing answer is not the same as proving that the answer is correct.
For now, the Navier–Stokes problem remains a test not only of mathematical ingenuity, but also of whether AI can produce discoveries that humans can independently understand and trust.
