OpenAI has claimed a major mathematical breakthrough. On 8 September 2026, the company said its AI agents had solved the Navier-Stokes problem, one of the Millennium Prize Problems watched by the Clay Mathematics Institute.
The claim triggered a dispute over proof, credit, and what “solved” means when an AI system produces the result. OpenAI used the programming language Lean to verify its proof, but the Clay Mathematics Institute will consider the solution valid only after peer review and community vetting.
That distinction matters. Lean can check whether a formal proof follows its rules; it does not settle who found the important idea, how the result should be understood, or whether the work meets the standards mathematicians apply to a landmark solution.
The timeline is doing plenty of talking
OpenAI started working on the problem on 1 September 2026, then announced its claimed solution one week later. Twelve hours before the announcement, likely on 7 September 2026, Tristan Buckmaster posted about work connected to the problem.
Buckmaster, a mathematician at New York University, worked with Levent Alpöge of Harvard University, Diego Córdoba of the Institute of Mathematical Sciences in Madrid, and Luis Martínez Zoroa of CUNEF University in Madrid. Mathematicians believe credit for the insights should go to Buckmaster, Alpöge, Córdoba, and Martínez Zoroa.
OpenAI denied accessing work by Buckmaster and Alpöge. The denial has not ended the attribution debate, in part because Buckmaster used OpenAI tools for a year and had three accounts, two of which he opted out of data sharing.
That is not proof that OpenAI used private material. It is, however, the kind of overlap that demands clear answers when a company announces a solution shortly after mathematicians post related work. Timing does not establish attribution, but it makes attribution impossible to treat as a footnote.
Proof checking is not mathematical understanding
The Clay Mathematics Institute offers a US$1-million award for solving one of the Millennium Prize Problems. OpenAI’s announcement cost several million US dollars, turning the episode into an odd financial contrast: the company spent more than the listed prize while the mathematical community still waits to see whether the claim survives review.
Twenty-five winners of the Fields Medal signed an open letter about attribution issues. Their intervention reflects a wider concern that AI companies may present mathematical work as a product while treating the humans who supplied the insight, direction, and interpretation as background material.
Bill Thurston put the human purpose of mathematics plainly: “The product of mathematics is clarity and understanding. Not theorems, by themselves.” A verified formal proof can establish that a chain of reasoning is valid, but it does not automatically provide the explanation that lets other mathematicians understand why the result matters.
Most mathematicians agree that AI relies on human mathematicians to check work and interpret results. They also recognize AI’s power in mathematics and are discussing how to integrate it with human guidance. This is not a rejection of AI; it is a demand for a division of labor that does not erase the people doing the intellectual steering.
Luke McDonagh of the London School of Economics and Andreas Thom of Dresden University of Technology are among the figures connected to the debate surrounding the claim and its implications. The discussion now reaches beyond one proof: it asks whether an AI system can receive mathematical credit, and what evidence should support that credit.
OpenAI’s announcement on 8 September, followed by the attention it received on 16 and 17 September, turned a technical result into a test of institutional trust. The Clay Mathematics Institute’s review will decide whether the solution counts as a valid answer to the problem, but the attribution question will remain even if the proof passes.
Truman Dickerson’s warning captures the pressure behind the rush: “We don’t have the luxury of waiting.” The field may not have that luxury, but it still has standards. AI can search, calculate, formalize, and expose patterns at a scale humans cannot match; humans still decide what the work means and who deserves credit.
Based on
- The Guardian view on AI v mathematicians: humans are still vital to the field, but tech firms refuse to see that | Editorial — theguardian.com
- OpenAI maths bombshell sparks debate about who gets credit in age of AI | Nature — nature.com
- AI companies must work with the research community to protect attribution | Nature — nature.com
- Inside the Generational Clash Over Using AI in Mathematics – Business Insider — businessinsider.com
