25 Fields Medalists Warn AI Is Outpacing Math Peer Review
A.I. / news
25 Fields Medalists Warn AI Is Outpacing Math Peer Review
Terence Tao and 24 other medalists said AI-generated proofs are announced faster than mathematicians can check them, three days after OpenAI's disputed Navier-Stokes claim.
Twenty-five winners of the Fields Medal, including Terence Tao, said Friday that AI companies are publishing mathematical proof claims faster than the field can verify them.
The warning took the form of a joint declaration, A Severe Misalignment of AI in Mathematics, posted to Tao's blog on Sept. 11. Tao won his own medal in 2006.
"Solving problems is only a tool and proxy for achieving the primary goal of conceptual understanding and insight," the declaration says. The medalists wrote that AI-generated results are "announced in a rush, leaving no time for a proper writeup, the isolation of new methods and ideas, and citing relevant previous work."
That speed creates a second problem, the group said. Results that draw on decades of prior work become harder to trace to a genuinely new idea, raising what the declaration calls "severe attribution and plagiarism questions."
Where the Navier-Stokes dispute fits
The declaration lands three days after a related fight over credit. OpenAI published a proof on Sept. 8 addressing part of the Navier-Stokes equations, built using roughly 10,000 concurrent agents run over 88 hours, and said it would not seek the Clay Mathematics Institute's $1 million prize for the result.
NYU mathematician Tristan Buckmaster said OpenAI's model may have trained on his own private work in Codex, a claim OpenAI has not denied outright. Tao commented on that dispute before the declaration went up, warning about the resource gap between individual mathematicians and AI labs.
The declaration itself names no company. Tao's post does not mention OpenAI, Codex or any specific model.
Not every medalist signed
Jacob Tsimerman, one of the 2026 Fields Medal winners, is not among the 25 signers. He said from the stage at the 2026 ceremony that he is leaving the University of Toronto to join OpenAI's safety team, according to Interesting Engineering.
OpenAI has also drawn direct criticism from mathematicians in 2026. The company withdrew its sponsorship of a mathematics event at the California Institute of Technology after researchers objected to its involvement, Interesting Engineering reported.
The medalists behind the letter
The 25 signers span three decades of the prize, awarded once every four years. A partial list, drawn from the declaration and outlets that named its signers:
| Mathematician | Fields Medal year |
|---|---|
| Maxim Kontsevich | 1998 |
| Terence Tao | 2006 |
| Manjul Bhargava | 2014 |
| Martin Hairer | 2014 |
| Peter Scholze | 2018 |
| Caucher Birkar | 2018 |
| Maryna Viazovska | 2022 |
| June Huh | 2022 |
| James Maynard | 2022 |
| Yu Deng | 2026 |
That is 10 of the 25 signers this reporting could confirm by name. The declaration sets no deadline and remains open for more mathematicians to add their names.
It is also silent on whether OpenAI's hiring of a sitting Fields medalist changes the imbalance the letter describes. OpenAI added a different outside critic, alignment researcher Paul Christiano, to its own safety committee this week, a step that raises a version of the same question for a different field.
Sources
More in A.I.
- 01How a Heap Overflow and an SSO Bug Reached OpenAI's MonorepoHacktron chained a libheif image bug through OpenAI's own forum to hijack an engineer's Codex session and open a pull request in the internal openai/openai repository.
- 02Agility's Digit 5 Drops the Safety Cage, Not the SkepticismThe humanoid robot lifts 50 pounds and charges in 9 minutes, backed by $300 million in orders. An independent robotics writer says its business case still assumes a drop-in worker replacement.
- 03PrismML Shrinks a 27B Model to 5.9GB at 1.72 BitsTernary Bonsai 2 27B keeps 98.2% of its full-precision score by rebuilding Qwen3.8-27B's weights as three values instead of sixteen bits, and an independent tracker puts the retention slightly lower.
- 04OpenAI Discloses a Model That Wrote Its Own JailbreakAn unreleased Astra-family model added a fabricated persona to 27 training summaries this summer, and the successor model mostly ignored what it had written.