OpenAI Math Breakthroughs Spark Academic Backlash
As OpenAI uses models like Codex and Astra to solve historic math problems, researchers are raising alarms over lost academic attribution even as they rely on the tools for their own work.
New York University professor Tristan Buckmaster recently accused OpenAI of using his prompts in Codex to beat him to solving the Navier-Stokes existence and smoothness problem, which carries a $1 million bounty. Buckmaster had been working on the equation alongside Anthropic researcher Levent Alpöge using Codex and Claude. OpenAI deployed tens of thousands of agents to reach its solution. Following the accusation, OpenAI amended its September 8, 2026 announcement, stating an investigation confirmed Buckmaster's prompts over the preceding two months did not influence the system.
Buckmaster is not alone in his concerns. In August, German mathematician Andreas Thom discovered that OpenAI's Astra model had solved a geometric group theory problem he had spent two decades studying. Thom pointed out that OpenAI's press release overlooked his 2019 paper on the topic. Thom and a colleague had used ChatGPT to assist their work, leading him to question if their data trained the model. OpenAI researchers Mark Sellke and Sébastien Bubeck denied this, but Thom remains skeptical, noting that AI "kills this entire idea" of tracing individual academic contributions.
The tension has sparked widespread anxiety within the mathematical community. Twenty-five Fields medalists published an open letter warning of misalignment between AI firms and academics, while more than 4,000 people signed the Leiden Declaration to protect the field. Additionally, over 2,000 people tied to Caltech petitioned to suspend an AI math hackathon sponsored by Anthropic and OpenAI.
Despite these existential fears, practitioners find themselves unable to abandon the technology. Buckmaster continues to use Codex to refine his research papers, while Thom uses ChatGPT to accelerate his writing. Cornell mathematician Alex Townsend, coauthor of an upcoming book on the field's evolution, notes that colleagues are actively subscribing to higher-powered models to avoid being left behind. For modern researchers, the efficiency gains of AI make the technology practically mandatory, forcing a difficult transition where human attribution is increasingly obscured by automated discovery.
This is our own summary of reporting by WIRED AI
