OpenAI's dispute with mathematicians deepens over unverified AI proofs

Twenty-five of the world's leading mathematicians, all recipients of the Fields Medal, have signed an open letter expressing concern over how artificial intelligence companies are conducting themselves in the field of mathematics. The signatories argue that AI labs, competing to outdo one another in solving famous math problems, are putting researchers' intellectual work at risk.

The dispute intensified this week when NYU professor Tristan Buckmaster accused OpenAI of pressuring him not to credit a collaborator working at Anthropic in connection with the solution to an important math problem. Buckmaster also suggested that OpenAI may have used their joint work involving the Codex tool to produce its own proof over an intense weekend of computation.

On Thursday, OpenAI withdrew its sponsorship of a mathematics event at Caltech after facing criticism from researchers at the institution.

Concerns over credit and verification

The letter's signatories argue that AI-generated math solutions can only benefit humanity if they can be properly understood and communicated by the mathematical community and the broader public. They criticize a pattern in which solutions are announced hastily, without time for proper writeups, identification of new methods, or citation of relevant prior work — noting that OpenAI's proof remains unverified. Without mathematicians willing to integrate these ideas into the broader body of mathematical knowledge, the letter states, AI-generated concepts would never be fully realized, and the essential chain of knowledge transmission between researchers would be broken.

Many mathematicians have grown wary, wondering whether their use of Codex could in turn be fed back into training OpenAI's newer models. There is genuine concern that frontier labs could spend tens of millions of dollars using large language models to beat original researchers to a proof, creating incentives toward secrecy rather than openness.

The letter follows the Leiden Declaration, published in June by a working group of mathematicians, which also addresses how AI-generated proofs are reshaping the field and offers recommendations for researchers, institutions, and policymakers. The signatories note that the challenges now facing mathematics mirror those confronting other scientific and creative fields, pointing to a broader question of how to preserve the purpose of work as AI changes how it gets done.