On September 11, 2026, 25 Fields Medal laureates published an open statement titled 'The Profound Mismatch of AI with the Goals of Mathematics,' warning that the rush by technology companies to use the solving of mathematical problems as a way to measure the performance of neural networks is harming mathematics itself. The document is also hosted on a separate website, mathandai.org.
The authors of the statement point out: in recent months, the mathematical capabilities of large language models have grown sharply—they are already solving major unsolved problems from various areas of the discipline. The problem lies elsewhere: the goals of companies developing artificial intelligence and the goals of the mathematical community are seriously diverging.
Why Problem-Solving Is Not the Main Goal
Research mathematics is concerned with understanding the fundamental structures of forms, numbers, and natural phenomena. Famous unsolved problems have traditionally served as benchmarks by which the community measures progress—but the solution itself has always been only an indirect indicator of the main goal: conceptual understanding. The result underwent a long journey through presentations, discussions, simplifications, and only then made its way into textbooks accessible to students.
If this logic is forgotten, a tool can turn against its own purpose. According to the signatories, the mass production of 'true/false' statements at ever-increasing rates is capable of destroying the soil from which new ideas are born, rather than enriching it.
A separate grievance concerns the rush to publish. AI-claimed results are often announced without proper description of the method, without highlighting new ideas, and without citing previous work, which raises questions of attribution and potential plagiarism. The authors of the letter emphasize: without the involvement of mathematicians who integrate findings into the overall canon of the discipline, AI products will never become full-fledged mathematical knowledge, and the chain of understanding transmission between people will be lost. Years of a researcher's training are needed not only to answer a specific question but also to develop the ability to formulate new questions—and this is precisely what is at risk of disappearing.
The Catalyst: OpenAI's Announcement on Navier-Stokes Equations
The statement appeared in the context of an announcement from OpenAI on September 8, 2026. The company reported that its internal system—a model significantly surpassing GPT-6 Astra—using 10,000 agents over 88 hours of work, found a proof that the Navier-Stokes equations can develop a singularity in finite time. OpenAI emphasized that it does not claim the Clay Mathematics Institute's Millennium Prize. Work on the problem began on September 1, 2026—following rumors of progress on one of the 'Millennium Problems'.
Concurrently, mathematician Tristan Buckmaster from New York University and Levent Alpoge from Anthropic obtained similar results for simpler equations, including forced Euler equations, using AI.
Who Signed the Statement
Among the 25 initial signatories are exclusively Fields Medal laureates from different years:
-
Artur Avila, 2014
-
Manjul Bhargava, 2014
-
Caucher Birkar, 2018
-
Pierre Deligne, 1978
-
Yu Deng, 2026
-
Simon Donaldson, 1986
-
Hugo Duminil-Copin, 2022
-
Alessio Figalli, 2018
-
Martin Hairer, 2014
-
June Huh, 2022
-
Maxim Kontsevich, 1998
-
Elon Lindenstrauss, 2010
-
Pierre-Louis Lions, 1994
-
James Maynard, 2022
-
Curtis McMullen, 1998
-
Shigefumi Mori, 1990
-
Ngô Bảo Châu, 2010
-
Andrei Okounkov, 2006
-
Peter Scholze, 2018
-
Stanislav Smirnov, 2010
-
Terence Tao, 2006
-
Maryna Viazovska, 2022
-
Cédric Villani, 2010
-
Wendelin Werner, 2006
-
Efim Zelmanov, 1994
The signatories do not deny the potential of AI to accelerate genuine mathematical understanding. But, as they emphasize, whether the changes bring benefit or a destructive effect depends on the decisions of the people controlling the technology, not on the algorithms themselves.
The authors of the statement insist on an urgent discussion of the raised issues—both within the mathematical community and with the participation of companies developing AI and society as a whole. For now, two pieces of news—OpenAI's proof regarding the Navier-Stokes equations and the letter from the 25 Fields Medal laureates—have highlighted an open contradiction between the pace of technological development and the traditional mechanisms of scientific verification.
AI Opinion
From the perspective of machine data analysis, the conflict appears broader than a single article. The rush around AI's mathematical achievements fits into the overall race among companies for the status of AGI creator: OpenAI's chief scientist, Mark Chen, in a parallel piece, estimated the company's progress as '80% of the way' to AGI, and the company's president, Greg Brockman, compared the current period to the historical moment of creating general intelligence. Such context suggests that solving the Navier-Stokes problem may have served not only science but also the corporate showcase for investors.
A technical aspect glossed over in the text is the question of verification: a proof constructed by ten thousand agents over 88 hours requires independent verification by human experts before it can enter the canon, and this process historically takes years, not days. The discrepancy between the speed of generation and the speed of verification is a risk capable of devaluing the achievements themselves. Will the pace of verification remain the same if the stream of 'proofs' only grows?
end-content
