Twenty-five Fields Medal recipients have signed an open letter warning that competition among AI laboratories is threatening mathematical credit, verification and the culture of open research. The intervention follows a dispute over OpenAI’s claim to a major proof and adds a coordinated response from some of the field’s most recognized scholars.
The letter does not reject AI-assisted mathematics. Its authors argue that machine-generated ideas become valuable only when mathematicians can understand them, document them and integrate them into the discipline’s accumulated knowledge.
A Proof Dispute Triggered Wider Concern
New York University professor Tristan Buckmaster accused OpenAI of pressuring him not to credit a collaborator employed by Anthropic for work on an important mathematics problem. He also questioned whether OpenAI used their research with Codex to create its own proof during an intensive weekend of model inference.
OpenAI’s proof remains unverified, according to TechCrunch. The company withdrew sponsorship from a mathematics event at Caltech on Thursday after researchers there criticized it. The source does not include OpenAI’s response to the new letter or explain the terms of the sponsorship decision.
The dispute builds on earlier reporting about competing claims and attribution. The new developments are the letter signed by 25 Fields Medalists and the withdrawn Caltech sponsorship, both of which move the conflict beyond two researchers into a broader institutional debate.
The Letter Defends Verification and Attribution
The signatories say AI solutions are often announced before there is a proper written proof, time to isolate new methods or a full record of related work. That pace can create plagiarism and attribution questions even when a generated result ultimately proves correct.
Mathematical validation involves more than checking a final answer. Researchers communicate why a method works, connect it with earlier ideas, teach it to others and identify new questions. The letter argues that AI-conceived ideas would not become part of the mathematical canon without that human transmission.
Some mathematicians also fear that their use of coding assistants could feed information into models that later compete with their own work. TechCrunch reports concern that wealthy frontier laboratories can spend tens of millions of dollars on inference once they identify a promising path, potentially reaching a proof before the researchers who opened that path.
The article does not establish that OpenAI trained on any particular mathematician’s private interaction or that such use occurred in this case. It reports a growing fear that uncertainty about data and credit could encourage researchers to keep work secret.
Mathematics May Preview Other Professions
The open letter follows the Leiden Declaration, published by a working group of mathematicians in June. That document also considers how language-model proofs will alter research and offers recommendations for scholars, institutions and policymakers.
Together, the statements frame mathematics as both a beneficiary and an early test of AI-mediated work. Faster proof generation could advance knowledge, while rushed announcements and uncertain attribution could weaken the collaborative structures that determine whether a result is meaningful.
The signatories extend the concern beyond their own field. They argue that scientific and creative professions face the same question: how to preserve the purpose and community around work when AI changes how outputs are produced. That is a normative position, not proof that every discipline will follow mathematics.
The immediate conflict remains unresolved. OpenAI’s result still requires verification, the allegations about pressure and credit have not been adjudicated, and the letter proposes principles rather than enforcement. Its significance lies in collective resistance from leading mathematicians who want AI-assisted discovery judged by the same standards of explanation, attribution and integration as human work.
Their demand is ultimately procedural: disclose the path, establish who contributed and allow enough time for the mathematical community to test what is being claimed.
