The Association for Human Mathematics has called for a boycott of OpenAI, prompted by the company's publication of research covering 377 mathematical problems alongside a repository holding upwards of 700 files. The organisation stated that mathematicians were never consulted before the work was undertaken, and contended that established norms governing scientific research were not followed by the company.

The disagreement unfolded against a backdrop of longstanding debate about what artificial intelligence can and should contribute to advanced mathematics. It follows OpenAI's announcement last month that an internal model had produced what the company described as a solution to the Navier-Stokes Millennium Prize problem, a long-standing challenge in the field.

Responding to OpenAI's October 6 publication, the Association for Human Mathematics took issue with how the company had conducted its AI-generated research, with the group stating that mathematicians had not been consulted on or invited to pursue that body of work. The organisation raised that point in direct contradiction of OpenAI's assertion that the findings would benefit the field of mathematics.

One argument put forward by the group held that releasing hundreds of documents reflected a bid for influence rather than a demonstration of genuine scholarly standards. The organisation further contended that OpenAI had set aside reservations put forward by the Advisory Group on Mathematics and Artificial Intelligence, a body that has advised against frontier AI developers using advanced mathematical problems to evaluate their own internal models. The organisation advised mathematicians and the wider public to treat the company's methodology with caution.

A September 29 statement from the Advisory Group on Mathematics and Artificial Intelligence — operating under the umbrella of Princeton's Institute for Advanced Study — put forward reservations about mathematical arguments produced by AI systems. The document noted that AI-generated reasoning processes may reach conclusions that those reviewing them find difficult to interpret, confirm, or formally take ownership of.

Concerns have also been raised that differential access to high-powered models and computing infrastructure — resources that independent researchers may not be able to obtain — could push mathematics toward a stratified landscape where some practitioners hold a distinct structural advantage over others.

The debate, as framed by those involved, extends beyond the question of whether AI can handle difficult problems. Questions of methodology — how findings come to be, how they are verified, and how they reach wider audiences — sit at the heart of the dispute, alongside the matter of whether all researchers can access the process on equal terms.

OpenAI's statement described a process in which the advisory group's input — both its formal recommendations and any public guidance — shaped the approach taken to publishing the latest research. Lean, a programming language whose built-in verification engine can confirm whether a mathematical proof holds, served as the vehicle for formal proof documentation when the organisation posted its findings to GitHub.

Among the materials OpenAI put forward were ten brief summaries of the model's reasoning process, alongside figures on how many problems the system attempted and approximations of the computational resources drawn upon. According to the findings, completing a typical task demanded processing power on par with roughly three hours of operation under ChatGPT Pro's thinking mode.

Those papers shed light on the research process itself, yet questions about how scientific work should be conducted, disclosed, and made accessible to the public remain unsettled. Underlying the disagreement is a question that researchers in artificial intelligence have yet to resolve: without independent access to the inner workings of the organisations responsible for both building and assessing these systems, how reliably reported advances in performance can be confirmed remains an open matter.