25 Fields Medalists Warn AI Labs Over Math Research and Attribution

A group of 25 Fields Medal-winning mathematicians warns that competition among AI companies could damage research norms around attribution, verification, and the open exchange of mathematical ideas.

In an open letter on mathematics and AI, the signatories argued that AI-generated solutions to major mathematical problems should not be rushed into public view without proper documentation, review and acknowledgement of earlier work.

The warning comes amid growing tension between OpenAI and parts of the mathematics community over how advanced AI systems are being used to pursue difficult proofs and how credit should be assigned when human researchers and AI tools overlap.

OpenAI faces questions over proofs and attribution

NYU professor Tristan Buckmaster this week accused OpenAI of pressuring him not to credit a collaborator employed by Anthropic for work on an important mathematical problem. He also questioned whetherresearchers” work with Codex may have contributed to OpenAI’s own subsequent effort on the problem.

OpenAI has separately published details of an AI-assisted Navier-Stokes solution, but the source report said the proof had not yet been independently verified.

The dispute widened Thursday when OpenAI withdrew sponsorship from a mathematics event at Caltech after researchers at the university criticised the company. The broader debate has also played out publicly among mathematicians, including in online discussions about OpenAI and mathematical research.

Mathematicians warn of pressure on open research

The Fields Medalists said AI-generated breakthroughs can be valuable only if mathematicians have enough time to understand, document and integrate new ideas into the wider body of mathematical knowledge.

“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,” the letter said.

The signatories also raised concerns about plagiarism and attribution when AI systems produce results based on existing mathematical work. They warned that researchers could become less willing to share preliminary ideas if well-funded AI labs can rapidly spend large amounts of computing resources pursuing the same proofs.

Those concerns build on the Leiden Declaration, released in June by a working group of mathematicians. That document offered recommendations for researchers, institutions and policymakers as large language models become more involved in mathematical discovery.

The latest letter argues that the value of mathematical research extends beyond producing a final proof. It also includes developing methods, identifying new questions, teaching future mathematicians and communicating ideas so they can become part of the broader mathematical record.

The signatories said the challenges now facing mathematics resemble those emerging across other scientific and creative professions as AI systems become more capable of participating in work traditionally performed by people.