25 Fields Medalists warn AI math benchmarks are misaligned with mathematics

Terence Tao published a joint declaration with 25 Fields Medallists as initial signatories, arguing that AI labs’ rush to solve hard problems as benchmarks clashes with mathematics’ goals of understanding, method, attribution, and teaching.

Twenty-five Fields Medal winners, via Terence Tao, say AI labs are treating famous math problems as a scoreboard — and that this clashes with how mathematics actually works. Their signed words are the news, not a product launch.

Terence Tao posted the joint declaration “A Severe Misalignment of AI in Mathematics” on his blog, naming himself among 25 initial signatories who are all Fields Medallists — winners of the field’s highest prize, awarded every four years. He says it grew from discussions over the prior week, that they also posted it at mathandai.org, and that they invite further signatures (similar to the Leiden declaration). That blog post is the filing event.

The declaration states that LLM mathematical capabilities — large language models, the same class of system as ChatGPT — have improved dramatically enough to solve major outstanding problems in many fields, but that AI companies’ push to solve mathematical problems as a benchmark is detrimental to the science of mathematics and the mathematical community. Quoted line on the page: “The goals of the AI companies and the goals of the mathematical community are severely misaligned.” Those are the signatories’ claims — not a desk audit of any lab’s methods.

Signatories argue that solving famous problems has historically been a proxy for new insights that then get studied, simplified, attributed, and taught. They warn that rushed AI “true/false” mass production can skip proper writeup, isolation of methods, citation of prior work, and the human transmission chain — raising attribution and plagiarism questions. Treat that as their position.

They frame the issue as part of broader alignment stress on scientific and creative professions: years of training have traditionally built understanding and the ability to ask new questions, while AI can increasingly produce end results directly. They say AI can enhance genuine mathematical study, but whether that helps or harms depends on human decisions about the technology.

Initial Fields Medallist signatories listed on Tao’s post: Artur Avila, Manjul Bhargava, Caucher Birkar, Pierre Deligne, Yu Deng (Fields Medal 2026), Simon Donaldson, Hugo Duminil-Copin, Alessio Figalli, Martin Hairer, June Huh, Maxim Kontsevich, Elon Lindenstrauss, Pierre-Louis Lions, James Maynard, Curt McMullen, Shigefumi Mori, Ngô Bảo Châu, Andrei Okounkov, Peter Scholze, Stanislav Smirnov, Terence Tao, Maryna Viazovska, Cédric Villani, Wendelin Werner, and Efim Zelmanov.