Glonce

25 Fields Medal winners warn AI benchmarks harm mathematics

researchThe Glonce editorsHacker News🔥78Glonce rating78Impact65Novelty55Relevance90Surprise45Players75Credibility85

Twenty-five mathematicians who have each won the Fields Medal, mathematics' highest honor, have published a joint open letter at mathandai.org titled 'A misalignment of AI in mathematics.' The signatories span Fields Medal years from Pierre Deligne (1978) to Yu Deng (2026), and include Simon Donaldson, Terence Tao, Peter Scholze and Maryna Viazovska among others. Their central claim is that large language models have improved so quickly over the last few months that they can now solve major outstanding problems in several fields of mathematics, but that the push by AI companies to treat that problem solving as a benchmark to clear is detrimental to mathematics as a science and to the mathematical community. The letter states plainly that the goals of the AI companies and the goals of the mathematical community are severely misaligned, and frames this as one visible case of a broader alignment failure it expects to also hit other scientific and creative professions, and society as a whole.

The letter describes research mathematics as a generations-long project of understanding the basic structures behind shapes, numbers and natural phenomena, in which famous unsolved problems have long served as landmarks for measuring progress. But solving one, the signatories argue, has traditionally been only the start: a community of mathematicians then spends a long, human-driven process of talks, discussion and simplification distilling the result, ideally ending in a textbook presentation accessible even to undergraduate students, with some ideas eventually reaching the wider population, understood and used by non-specialists, decades or centuries later. They describe the mathematical community itself as functioning like a miniature version of humanity, joined by shared values around what the letter calls its most precious resources, students and ideas, both nurtured with care so they can grow to their full potential and eventually stand on their own in the mathematical world.

Against that backdrop, the letter raises three specific worries about AI-produced mathematics. First, treating problem solving as only a benchmark forgets that it is a tool for conceptual understanding rather than the goal itself, and mass-producing 'true/false' statements at a faster and faster pace could destroy the field's fertile ground rather than breathe life into it. Second, such solutions are often announced in a rush, without a proper writeup, without isolating the new methods involved, and without citing relevant prior work, which the letter says raises severe attribution and plagiarism questions, though it stops short of saying plagiarism has actually occurred. Third, without mathematicians willing to develop these ideas and integrate them into the field's canon, the letter argues they would never become fully alive, breaking the human transmission chain that has historically passed mathematical understanding from one mathematician to the next.

The signatories place this inside a larger pattern: years of training and work across many fields have traditionally served not only to produce a final answer or product, but also to build understanding and the ability to formulate new questions, and AI systems, built on a vast body of previous human work, are becoming increasingly capable of producing the results of that work directly, so the two goals cease to align. They do not reject a role for AI in mathematics: the letter states that AI offers real potential to enhance and accelerate genuine mathematical study, and that the profession will need to adapt in several ways. But it says whether that turns out well or badly will, in large part, be determined by the decisions of whoever controls the technology, and it calls for the issue to be addressed urgently by the mathematical community, by the companies developing these technologies, and by a society the letter expects to face similar problems in other forms of intellectual work.

Key facts

  • Twenty-five mathematicians who have each won the Fields Medal, spanning laureates from Pierre Deligne (1978) to Yu Deng (2026), signed a joint open letter published at mathandai.org and titled 'A misalignment of AI in mathematics.'
  • The letter states that the push by AI companies to solve mathematical problems as a benchmark is detrimental to mathematics as a science and to the mathematical community, and that the goals of the AI companies and the mathematical community are severely misaligned.
  • It warns that AI-produced solutions are often announced without a proper writeup or citation of prior work, which raises severe attribution and plagiarism questions, and that without mathematicians integrating such results into the field's canon, the ideas would never become fully alive and the human transmission chain between mathematicians would be lost.
  • The signatories frame this as one case of a broader threat to intellectual work, saying whether AI's effect on mathematics turns out beneficial or destructive will, in large part, be determined by the decisions of whoever controls the technology.
  • The letter names no specific AI company, model or already-solved problem and proposes no concrete remedy beyond an urgent call to the mathematical community, AI companies and society; on Hacker News it drew 777 points and 767 comments within about 11 hours of posting.

Why it matters

That 25 of mathematics' most decorated researchers, spanning Fields Medal years from 1978 to 2026, put their names to the same public letter is itself unusual. Their argument is that large language models have grown capable enough, over just the last few months, to solve major outstanding problems in several fields of mathematics, but that AI companies treating that as a benchmark to clear is detrimental to mathematics as a science and to the mathematical community, because it optimizes for producing verified true or false statements rather than for the understanding, methods and teaching a real mathematical advance is supposed to generate. The letter states plainly that the goals of the AI companies and the goals of the mathematical community are now severely misaligned, and frames mathematics as an early, visible case of a broader alignment problem it expects to spread to other scientific and creative professions, and eventually to society at large.

Who it affects

Working mathematicians and their students most directly: the letter describes problems typically being chosen for students to build skills that position them for future research, not just to produce an answer, and frames the field's normal path via talks, private discussions and careful writeups as depending on human interaction. It also names the wider mathematical community, which studies, simplifies and eventually teaches new results, and the AI companies building systems capable of tackling research-level problems, referred to only generically since the letter names no specific company or model. The signatories close by addressing society more broadly, saying they expect other scientific and creative professions, and other forms of intellectual work generally, to face the same pattern they describe in mathematics.

How to use it

This is a position statement, not a tool or a policy, so there is nothing to install or adopt from it. Its practical content is a three-way call: it asks the mathematical community, the companies building these AI systems and society more broadly to address the misalignment urgently. The letter does not reject AI's role in mathematics: it explicitly says AI offers real potential to enhance and accelerate genuine mathematical study and understanding, and that the profession will need to adapt in several ways. But it gives no concrete mechanism, deadline or policy for doing so. The closest thing to practical guidance is a standard for judging AI-produced mathematics: whether a result gets a proper writeup, clear attribution and eventual integration into how the field teaches, not just whether a benchmark got cleared.

How solid is it

The letter is posted at a dedicated website, mathandai.org, rather than run through a journal or a news outlet, and it names all 25 signatories individually alongside their Fields Medal year, from Pierre Deligne (1978) to Yu Deng (2026), including widely known figures such as Simon Donaldson, Terence Tao, Peter Scholze and Maryna Viazovska. It is written throughout in a collective 'we' voice: no individual signer is quoted or presented as its author or spokesperson, and the text says nothing about how the group was assembled or whether anyone approached declined to sign. It names no specific AI company, model or already-solved problem, so its central claim of 'severe' misalignment cannot be checked against a concrete case from the letter itself. Reception has been substantial: on Hacker News, where the letter surfaced, discussion reached 777 points and 767 comments within roughly 11 hours of posting.

Risks and caveats

This is an opinion and values statement from a group of researchers about their own field, not a study backed by data: it names no AI company, model or solved theorem, so its claims cannot be checked against a specific case. Its wording on misconduct is more careful than a summary of it might suggest: the letter says rushed, uncredited AI solutions 'raise' attribution and plagiarism questions, not that plagiarism has occurred. It similarly hedges its own conclusion, saying AI's ultimate effect on mathematics will be determined 'in large part', not entirely, by the decisions of whoever controls the technology, leaving room for other factors it does not name. No timeline, mechanism or concrete policy accompanies its call to address the issue urgently, and no publication date appears anywhere in the text.

“The goals of the AI companies and the goals of the mathematical community are severely misaligned.”

— the open letter