Twenty-five Fields Medal winners, essentially a large share of the living mathematicians who have ever been awarded the highest honour in the field, have signed onto a joint declaration warning that the way AI companies are racing to solve famous mathematical problems is actively damaging the discipline. The statement, titled “A Severe Misalignment of AI in Mathematics,” was published on Terence Tao’s blog on September 11, and describes the goals of AI labs and the goals of the mathematical community as “severely misaligned.”
The signatories are a who’s who of modern mathematics: Terence Tao, Peter Scholze, Maryna Viazovska, Caucher Birkar, June Huh, Martin Hairer, Maxim Kontsevich, James Maynard, Manjul Bhargava, and 2026 medalist Yu Deng among them, spanning Fields Medal classes from 1978 all the way to this year’s ceremony. According to the group, the statement grew out of a week of discussions between the mathematicians themselves, and they decided the situation was urgent enough that they released it without the more consultative process used for similar declarations in other fields.
What the mathematicians are actually objecting to
The declaration doesn’t dispute that AI has gotten dramatically better at mathematics. It says the opposite: that LLM capabilities have “improved dramatically” over the past few months, to the point where they can solve major outstanding problems. That’s precisely the problem, in the mathematicians’ framing. Solving a famous open problem has traditionally been a signal, a “landmark” that pointed the community toward genuinely new ideas, which would then be studied, simplified, taught, and eventually absorbed into the canon over years of human effort. The declaration argues that AI companies, chasing benchmarks and headlines, are turning problem-solving into a “mass production” of true/false statements that displaces this process rather than feeding it.
The mathematicians single out the pace of these announcements as part of the harm. Solutions are being announced “in a rush,” they write, leaving no time for a proper writeup, no isolation of what’s actually new, and, they say, no accounting of prior work that the AI-generated proof may be drawing on. That last point reads as a direct response to the controversy that broke out earlier, when mathematicians accused OpenAI’s Astra system of reusing published results without proper credit in the math advances it announced.
It’s also hard to read the declaration without recalling this year’s Fields Medal ceremony itself, when 2026 winner Jacob Tsimerman announced from the stage that he was leaving the University of Toronto to join OpenAI’s safety team, arguing AI would soon do everything mathematicians do “better and faster.” Tsimerman isn’t one of the 25 signatories, and his framing — that mathematicians should get inside the labs shaping these systems rather than watch from the sidelines — sits in some tension with the declaration’s message that the labs’ current incentives are actively corrosive to the field.
A discipline built on people, not just answers
Much of the declaration reads less like a technical complaint and more like a defense of what mathematics is actually for. The signatories describe the field as “a miniature version of humanity,” where the most precious resources are students and ideas, nurtured through talks, private discussions, and careful writeups that connect new work to what came before. Problems given to students, they write, are chosen deliberately to build skills, not just to produce answers.
Their fear is that when an AI can produce the “results” of decades of training directly, it severs a transmission chain that has nothing to do with correctness and everything to do with how the human community actually grows and passes on understanding. As they put it, without mathematicians willing to develop and integrate AI-generated ideas into the canon, those ideas “would never become fully alive.”
This isn’t a new theme for Tao specifically. He has used ChatGPT and the Lean proof assistant together to formalize proofs, describing the AI as circling back on itself and citing the wrong theorems until nudged, more a junior collaborator that needs supervision than a peer. That contrasts sharply with the tone of this new declaration, which is aimed less at individual tools and more at the industry-wide incentive structure around them.
Part of a bigger pattern in 2026
The declaration arrives in a year that has been very eventful for AI and mathematics. In July, Anthropic researcher Levent Alpöge used the company’s Claude Fable 5 model to find a counterexample disproving the 87-year-old Jacobian conjecture, announcing it on social media the night of the World Cup final in a post that became one of the year’s most talked-about AI moments. Weeks later, OpenAI said roughly 10,000 concurrent agents had solved a Navier–Stokes problem in about 88 hours at a cost in the millions of dollars — a claim that drew its own controversy after NYU mathematician Tristan Buckmaster, who had been working on closely related research with Alpöge, questioned whether OpenAI’s effort had been influenced by their unpublished work. OpenAI said no people or systems had seen it beforehand.
Taken together, the Fields Medalists’ declaration reads as the mathematics community’s attempt to draw a line under all of it: acknowledging that the capability jump is real, while insisting that racing to claim credit for it, without the years of verification, writeup, and community absorption that mathematics has always depended on, is doing more harm than good.
The mathematicians frame their concern as bigger than their own field. “The issues the mathematical community faces now are similar to issues that other scientific and creative professions are facing,” they write, “and indicate issues that all of humanity might face.” The declaration is open for further signatures at mathandai.org.