What the declaration says
Twenty-five winners of the Fields Medal have signed a statement titled “A Severe Misalignment of AI in Mathematics”, published on Terence Tao’s blog on 11 September. Its central claim is blunt: “The push by AI companies to solve mathematical problems as a benchmark is detrimental to the science of mathematics, and to the mathematical community.”
The signatories do not argue that machine-produced proofs are worthless. They argue that a proof is not the point. Solving problems, the statement says, “is only a tool and proxy for achieving the primary goal of conceptual understanding and insight” — and that the goal is what the current announcement cycle skips.
The specific complaint is about speed
The mechanism the mathematicians describe is procedural rather than philosophical. “Often these solutions are announced in a rush, leaving no time for a proper writeup, the isolation of new methods and ideas, and citing relevant previous work,” the declaration says. That rush, they write, raises severe attribution and plagiarism questions.
There is a second-order problem behind it. A result enters mathematics when other mathematicians read it, check it, generalise it and teach it. The statement puts it this way: without “the willing mathematicians who must take care of their development and integration into the mathematical canon, AI-conceived ideas would never become fully alive”. A community that feels cut out of the process is less likely to do that work.
Who signed
The signatories span Fields Medal classes from 1978 to this year’s ceremony, and include Terence Tao, Pierre Deligne, Maxim Kontsevich, Peter Scholze, Maryna Viazovska, Caucher Birkar, June Huh, Martin Hairer, James Maynard, Manjul Bhargava and 2026 medallist Yu Deng. For a prize awarded to at most four people every four years, twenty-five names is close to the whole living field of recent laureates.
The month that produced it
The declaration lands after a fortnight of open friction. On 8 September, NYU professor Tristan Buckmaster alleged that OpenAI had pressured him not to credit an Anthropic collaborator for solving an important problem, and questioned whether the company had drawn on Codex work in developing its own proof, TechCrunch reported. On 11 September, the same day the statement appeared, OpenAI withdrew funding from a mathematics event at CalTech after university researchers criticised the company.
That follows a run of announcements in which AI labs claimed results on long-standing problems, including OpenAI’s work on the Navier-Stokes equations and Anthropic’s formalisation of Fermat’s Last Theorem.
What they are asking for
The asks are concrete and, notably, procedural: responsible disclosure practices from AI companies, proper citation of prior work, results developed far enough that human mathematicians can absorb them, and what the statement calls an urgent dialogue between the mathematical community, technology developers and society.
None of that requires a lab to stop working on mathematics. It does require the announcement to arrive with a write-up attached, which is a slower cadence than a benchmark scoreboard rewards. The test of whether this changes anything is the next claimed result: whether it is published with citations and a paper, or posted first and documented later.
