Fields Medal Winners Say AI Labs Are Breaking How Mathematics Works
Twenty-five of the world's top mathematicians have signed an open letter saying the race between AI companies to crack famous unsolved problems is undermining how math actually works, and who gets credit for doing it.
Key points
- Fields Medal winners, the mathematicians awarded the highest prize in their field, signed an open letter warning that AI labs are threatening the culture of open mathematical research.
- NYU professor Tristan Buckmaster accused OpenAI of pressuring him not to credit an Anthropic collaborator for solving an important math problem.
- OpenAI withdrew its sponsorship of a CalTech math event on Thursday after researchers at the university publicly criticised the company.
- The letter warns that OpenAI's high-profile proof, produced over a single weekend, remains unverified.
- The letter follows the Leiden Declaration, a separate set of recommendations for mathematicians and policymakers released by a working group in June.
The letter is blunt. Fields Medal winners, each holding the closest thing mathematics has to a Nobel Prize, published an open letter this week saying AI labs are crashing their field and leaving a mess behind.
The trigger was a series of disputes over credit and conduct involving OpenAI. First, as reported by TechCrunch AI, NYU professor Tristan Buckmaster publicly accused OpenAI of pressuring him to leave out a collaborator's name from a key math result. That collaborator works for Anthropic, a rival AI company. Buckmaster also questioned whether OpenAI had used researchers' own work with Codex, an AI coding tool, to fuel a headline-grabbing proof the company produced in a single marathon weekend of computation. We first reported on the credit dispute around that proof on 11 August, in "OpenAI's AI Solved 10 Hard Maths Problems. Now Mathematicians Are Asking What Their Jobs Are Worth.".
Then, on Thursday, OpenAI pulled its sponsorship from a math event at CalTech after researchers there criticised the company.
What exactly are the mathematicians worried about?
Three things stand out: speed, secrecy, lost credit. AI labs can spend tens of millions of dollars running large language models, the same technology that powers chatbots like ChatGPT, to chase solutions to famous unsolved problems faster than the human researchers who spent years working toward them.
The signatories say that race produces sloppy results. "Often these solutions are announced in a rush, leaving no time for a proper writeup, the isolation of new methods and ideas, and citing relevant previous work of others," they wrote. OpenAI's own recent proof, they note, has not yet been verified by the wider mathematical community.
Beyond attribution, there's a structural fear. If frontier labs routinely beat researchers to discoveries, working mathematicians may stop sharing their methods openly, afraid that doing so only helps a company beat them to the finish line.
| Document | Who published it | When | Focus |
|---|---|---|---|
| Fields Medal open letter | Fields Medal winners | This week | Credit, verification, open research |
| Leiden Declaration | Working group of mathematicians | June | Practical recommendations for institutions and policymakers |
| OpenAI proof | OpenAI | This week | Unverified mathematical result, produced over one weekend |
The Leiden Declaration, released in June and covered by AI2Day on 20 August, covered similar ground with a calmer tone, offering concrete recommendations for universities and governments. This new letter is sharper.
Should people outside mathematics care?
Yes, and the letter says so plainly. "The issues the mathematical community faces now are similar to issues that other scientific and creative professions are facing," the signatories wrote. When AI does the work, what's lost if no one can explain how the answer was reached, or if the people who laid the groundwork get no credit at all?
Mathematics may be first because it's measurable. A proof is either right or wrong, which makes it a clean target for AI benchmarks. Other fields where knowledge builds slowly on publicly shared research are watching closely.
My read: the core complaint here isn't really about AI solving hard problems. It's about a funding and incentive structure that rewards announcing a solution over understanding one. That problem existed before AI; the compute budgets just made it faster and harder to ignore.
Common questions
What is a Fields Medal, and why does it matter who signed this letter?
The Fields Medal is widely considered the most prestigious honour in mathematics. These signatories aren't outsiders complaining about technology; they're the people whose work AI companies are racing to replicate.
Does an AI proof count if mathematicians haven't checked it?
Not yet, and that's the point. In mathematics, a proof only enters the shared body of knowledge once other experts have read every step and agreed it holds. OpenAI's recent result hasn't gone through that process, which means it's an unverified claim, not an established fact.
