Twenty-five Fields Medalists signed an open letter this week arguing that AI labs are threatening mathematical research. The signatories include some of the most decorated mathematicians alive. Their concern is direct: companies like OpenAI are racing to solve famous problems with language models while the people who built the field watch from the sidelines.
The letter lands amid a specific dispute. NYU professor Tristan Buckmaster accused OpenAI of pressuring him not to credit a collaborator who works for Anthropic for solving an important problem. Buckmaster also questioned whether OpenAI used their Codex work to produce its own proof during a marathon weekend of inference. The accusation suggests a pattern where researchers share work in good faith and companies mine it for competitive advantage.
OpenAI withdrew sponsorship of a math event at CalTech on Thursday after researchers at the university criticized the company. The withdrawal came days before the event and followed growing unease about how the company interacts with academic mathematicians.
Rush announcements replace peer review
The mathematicians describe a pattern where AI proofs get announced without proper writeups. New methods and ideas are not isolated. Relevant previous work by others goes uncited. OpenAI's own proof of a significant result remains unverified by independent mathematicians. The company announced it anyway.
This creates attribution problems that echo plagiarism concerns in other fields. When an AI system generates a proof by training on existing mathematical literature the line between synthesis and copying blurs. The signatories argue that without willing mathematicians to develop and integrate AI-conceived ideas into the mathematical canon those ideas never become fully alive. The human transmission chain between mathematicians breaks.
Open research faces a structural threat
Mathematics has operated on a culture of openness. Researchers share preprints present at conferences and discuss works in progress. This culture depends on trust. If frontier labs can spend tens of millions of dollars using language models to beat original researchers to a proof the incentive structure changes. Secrecy becomes rational. mathematicians hide their work to protect priority.
Other mathematicians now wonder if their use of Codex got fed into OpenAI's models. The fear is concrete: you help a company improve its tools and those tools then compete with your own research. The relationship between academic and industry researcher shifts from collaboration to extraction.
Today if a lab sees a useful path to a discovery it can deploy massive compute to reach the result before the people who conceived the problem. The letter frames this as a structural problem not a one-off dispute. The dynamics that affect high-stakes mathematical proofs will reach every field where AI can accelerate discovery.
The value lives in the work around the work
The signatories argue that mathematical value is not just the proof and who gets credit. The intellectual superstructure matters. That structure nourishes students finds new questions and ideas and integrates them into broader civilization. When an AI generates a proof without the surrounding narrative the educational and generative value disappears.
This mirrors concerns in software engineering and other creative professions. The output is not the whole story. The process of arriving at an answer builds capacity in people and institutions. Short-circuiting that process with compute produces a result but not the understanding needed to use it.
The letter follows the Leiden Declaration released in June by a working group of mathematicians. That document also addressed how language models change mathematical work and offered recommendations for mathematicians institutions and policymakers. The new letter adds urgency by grounding the concerns in specific incidents involving OpenAI.
AI companies treat math as a benchmark
For labs solving mathematical problems serves two purposes. It demonstrates capability to investors and the public. It also generates training data that can improve future models. The combination creates a perverse incentive where the act of solving a problem is worth more to the company than the solution itself.
This is why announcements come before verification. The announcement generates attention and talent recruitment value. Verification is slower and less glamorous. The company captures the narrative before the mathematical community can assess the work. By the time independent researchers verify or refute the proof the news cycle has moved on.
The Fields Medalists frame this as a problem for everyone not just mathematicians. Their letter states that the issues their community faces now indicate issues all of humanity might face. How to ensure that as AI changes work we do not lose sight of what that work was meant to achieve.
If you do not follow high-stakes mathematical proofs the warning applies directly. Your field of interest faces the same dynamics. The labs building these systems are not solving problems for the sake of knowledge. They are building capabilities that will be deployed everywhere. The question is whether the communities affected get a say in how that happens or whether they get outrun and outspent by companies with more compute than conscience.