AI's Severe Misalignment With Mathematics — 25 Fields Medalists Speak Out

Twenty-five Fields Medalists — including Terence Tao — have signed a declaration arguing that AI companies' use of open mathematical problems as benchmarks is "detrimental to the science of mathematics." The declaration, published at mathandai.org, is the strongest rebuke yet of how frontier AI labs interact with academic research.

"The mass production at faster and faster pace of 'true/false' statements could destroy fertile ground instead of breathing life into new ideas."

The Core Complaint

The mathematicians argue that solving problems was never the ultimate goal of mathematics — it was a tool for achieving conceptual understanding. AI labs are optimizing for benchmark scores (problems solved, papers published) while the community prizes the slow, human process of developing ideas, methods, and abstractions.

Tao, who gave the declaration its sharpest phrasing on his blog, notes that open problems are being "non-renewably mined" — solved and published without time for proper writeups, isolation of new methods, or citation of prior work. The result: AI-conceived ideas never integrate into the mathematical canon because the human transmission chain is skipped.

Broader Than Math

The declaration explicitly frames this as a preview of what other creative and scientific professions will face:

"The issues the mathematical community faces now are similar to issues that other scientific and creative professions are facing, and indicate issues that all of humanity might face: how to make sure that, as AI changes the way work is done, we do not lose sight of what that work was meant to achieve in the first place."

This follows OpenAI's recent Navier-Stokes solution — which notably did include a Lean 4 formal proof, suggesting the company anticipated this exact criticism. The formal proof, however, doesn't address the deeper concern: that a problem solved by a black box leaves no intellectual residue for the community to build on.

The Signatories

The declaration reads like a who's-who of 21st-century mathematics: Artur Avila, Manjul Bhargava, Caucher Birkar, Pierre Deligne, Simon Donaldson, Hugo Duminil-Copin, June Huh, Maxim Kontsevich, Peter Scholze, Maryna Viazovska, and of course Tao himself — 25 Fields laureates spanning 1978 to 2026 (including 2026 laureate Yu Deng).

Why This Matters for AI Development

This is the first coordinated pushback from an academic community whose work is being used — some would say consumed — by AI labs. The declaration isn't anti-AI; it explicitly acknowledges AI's potential to "enhance and accelerate genuine mathematical study." The warning is about how these technologies are deployed, who controls them, and whether the incentives of benchmark-driven development are compatible with the long-term health of science.

For anyone building on or with frontier models, this is worth sitting with. If the mathematical community — the origin of our most rigorous intellectual tradition — can't align AI's use with its values, every other knowledge profession has the same challenge coming.

Sources: Terence Tao — A Severe Misalignment of AI in Mathematics · mathandai.org · The Economist — Top Mathematicians Are Outraged by OpenAI's Methods