Two dozen Fields Medalists warn AI's math-benchmark race endangers mathematics

An open letter signed by 24 Fields Medalists — including Terence Tao, Peter Scholze, and Maryna Viazovska — argues that the AI industry’s rush to crack famous mathematical problems as benchmarks is fundamentally at odds with what mathematics is for. LLMs have recently grown capable enough to settle long-open problems, but the signatories stress that solving a problem is only a proxy for the real goal: conceptual understanding and insight. Treating proof-generation as a scoreboard, they warn, risks mass-producing true/false verdicts while eroding the slow human process — talks, discussions, simplifications, textbook writeups — through which results actually become knowledge.

The letter raises concrete professional harms. Rushed AI-driven announcements often skip proper writeups, fail to isolate new methods, and neglect to cite prior work, creating serious attribution and plagiarism problems. More structurally, mathematical ideas only ‘come alive’ when human mathematicians integrate them into the canon and pass them to students; automating the output threatens to break that transmission chain and the training pipeline that develops future researchers’ ability to ask new questions.

The mathematicians frame this as a specific instance of a broad AI-alignment problem facing all intellectual and creative work: as systems learn to deliver the end product directly, the purpose that the work was meant to serve — understanding, skill, the capacity to formulate new ideas — can be lost. They call for urgent action from the mathematical community, the AI companies building these tools, and society at large, noting that whether AI ultimately helps or damages the field depends on the choices of the humans who control the technology.