Mathematicians warn AI labs over attribution and research openness

A group of 25 leading mathematicians, all recipients of the Fields Medal, signed an open letter this week stating that artificial-intelligence laboratories pose risks to the integrity of mathematical research. The signatories argue that rapid AI-generated proofs are often announced without thorough documentation, proper citation of prior work, or peer-reviewed exposition, which hampers their integration into the broader mathematical literature.

The letter cites a recent incident in which NYU professor Tristan Buckmaster said OpenAI allegedly discouraged him from acknowledging a collaborator from Anthropic after the latter contributed to solving a notable problem. Buckmaster also raised the possibility that OpenAI used work produced with its Codex system to generate a new proof over a short period of intensive inference.

OpenAI responded to criticism from CalTech researchers by ending its sponsorship of a mathematics event at the institute. The company’s recent proof remains unverified, prompting the mathematicians to highlight attribution and plagiarism concerns, and to stress the importance of human oversight in developing and disseminating new mathematical ideas.

The signatories warn that the current environment could encourage secrecy, as research groups might invest large sums to use large language models to achieve discoveries before the original investigators can publish. They note that similar challenges are emerging in other scientific and creative fields, and call for guidelines to preserve the collaborative and educational functions of mathematics.

The letter follows the Leiden Declaration released in June, which offers recommendations for mathematicians, academic institutions and policymakers on how to address the impact of AI-generated proofs on the discipline.