Wired reports growing tension between mathematicians and leading AI companies as OpenAI prepares to release results for more than 100 long-standing mathematical problems. The dispute is less about AI’s expanding mathematical capabilities than about the way discoveries are published, verified, and credited.
In August, OpenAI gathered about 40 mathematicians to discuss a future in which AI could outperform human researchers. Attendees learned that OpenAI’s models had made progress on numerous unresolved problems. Mathematicians advised the company to publish detailed papers rather than announcing results through blogs, social media, or repositories, allowing researchers to examine methods, verify proofs, and understand connections to existing work.
The issue became more contentious following OpenAI’s work on the Navier-Stokes Millennium Prize problem. New York University mathematician Tristan Buckmaster accused the company of racing ahead after learning about related unpublished work he had conducted with Anthropic researcher Levent Alpöge. OpenAI disputes aspects of the allegations. The episode intensified concerns about attribution, competition between AI companies, and the concentration of research power.
OpenAI formed an advisory group in September to guide its evaluation and communication of mathematical discoveries. Still, several mathematicians interviewed by Wired argue that AI companies continue to bypass established academic practices. Researchers have responded by creating tools such as Hexagon, a repository for primarily AI-generated mathematics, and Palomar, a registry for machine-verified results. More than 4,000 mathematicians have also backed the Leiden Declaration, which calls for responsible standards in AI-assisted mathematics.
Despite these concerns, mathematicians are not rejecting AI. Many see increasingly capable models as valuable research tools that could help tackle more ambitious problems. The central challenge is establishing scientific practices that keep AI-generated mathematics transparent, verifiable, properly attributed, and useful to the broader research community. The debate signals a deeper transition as AI begins influencing not only mathematical computation but also the process through which discoveries enter the scientific record.
