Image: METAL
Summary
- Twenty-five Fields medalists are the first signatories of a declaration titled A Severe Misalignment of AI in Mathematics. The list runs from Pierre Deligne, who won in 1978, to Yu Deng, who won in 2026.
- The declaration says the way AI companies push problem-solving as a performance benchmark is harmful to the discipline and its community. It also argues that rushed announcements leave no time for citation or proper write-up.
- It does not reject AI. While acknowledging the potential to widen research, it insists that whether the outcome is beneficial or destructive rests on the decisions of the humans holding the technology.
Mathematicians have never worried about how fast problems get solved. Spending years on a single problem is normal in the field, and the long process of refining a solution through talks, discussion and rewriting until it finally settles into a textbook was the discipline itself. A document written by mathematicians says that rhythm has broken over the past few months. It is titled A Severe Misalignment of AI in Mathematics.
The first signatories at the foot of the document are 25 Fields medalists. Nearly half a century of laureates is gathered on one page, from Pierre Deligne, who won in 1978, to Yu Deng, who won in 2026. Terence Tao, Peter Scholze, Maxim Kontsevich, Cédric Villani and June Huh also put their names to a document that takes direct aim at AI companies competing to crack mathematical problems. METAL checked the signature page, which as of the morning of September 12 carried 1,687 names, and a signature is added to the list only after identity is confirmed through an ORCID researcher identifier or an institutional email address. Professor Tony Carbery of the University of Edinburgh is one of a long run of working researchers, alongside names from the Massachusetts Institute of Technology, Northwestern University and ETH Zurich.
The claim compresses into a single sentence. The declaration states that the goals of the AI companies and the goals of the mathematical community are severely misaligned. Over the past few months, language models have improved enough in mathematics to solve long-standing open problems across several fields, and the signatories argue that the way AI companies push that problem-solving as a performance benchmark is harmful to mathematics as a discipline and to its community. They see the misalignment as part of a problem facing other sciences, creative professions, and ultimately society as a whole.
The heart of the argument is that solving problems was never the goal. Famous problems, the declaration explains, have served as lighthouses for measuring the mathematical landscape, and solving one was a reliable sign that new insight and new methods had arrived. A solution is only a tool and a proxy for the primary goal of conceptual understanding, and forgetting that in the world of AI turns the tool against the goal. Mass-producing true-or-false statements at an ever-faster pace, they wrote, could destroy fertile ground rather than breathe life into new ideas. It is the worry of an orchard picked clean for its fruit, with no trees left to bear the next crop.
The criticism of how results are announced is more specific. The declaration points out that solutions are published in a rush, leaving no time for a proper write-up, for isolating the new methods, or for citing the earlier work of others. It adds that, as in all creative professions, this raises severe attribution and plagiarism questions. Results are piling up while the question of whose contribution counts, and how far, goes unsettled.
What they call most precious is neither the problems nor the solutions. The declaration says the most precious resources of the profession are students and ideas, and that these are nurtured with great care. Problems are suggested to students not to obtain answers but to build the capacity to move forward in research and beyond. Without willing people to tend them, ideas conceived by AI never fully come alive, and the human chain of transmission between mathematicians is lost. That is their warning.
The declaration pushes the problem out past mathematics. In many fields, years of training served not only to produce a final result but to develop understanding and the ability to pose new questions, and as AI, building on a vast body of prior human work, begins delivering those results directly, the two purposes are coming apart. The signatories use the phrase general threat to intellectual work without hedging it. The question is how to keep sight of what the work was meant to achieve as the way it is done changes.
The declaration does not call for AI to be thrown out. It says AI has the potential to enhance and accelerate genuine mathematical study and understanding, and accepts that mathematics as a profession will need to adapt in several ways. But whether those changes benefit the field or prove destructive, it insists, will in large part be determined by the decisions of the humans in control of this new technology. It is a sentence that puts responsibility back on the people holding the technology rather than on the technology.
This concern did not arrive out of nowhere. METAL reported on Terence Tao's remark that good mathematical problems are being non-renewably mined toward exhaustion, and Tao is among the first signatories here. METAL has also covered a machine finishing the verification work on a major proof that used to take years. In a matter of months the axis of the argument has moved from whether machines can solve to who receives the solution and how.
What marks this declaration is how clearly it names its audience. The signatories write that the issues must be addressed urgently, not only within the mathematical community but by the companies developing the technology and by society at large, and signatures are accepted only from people who pass an academic identity check. One statement will not halt the benchmark race, but the fact remains that the community expected to read, verify and carry forward these solutions has now put its objection in writing. A mathematical problem is now priced not at the moment it is solved but at the moment it is taken up.
Comments