NYU Mathematician Alleges OpenAI Misused His Work in Race to Solve Clay Problem

NYU mathematics professor Tristan Buckmaster announced over the weekend that he and Anthropic mathematician Levent Alpöge have produced three proofs, including a preliminary solution to a major unsolved problem in theoretical mathematics. However, the announcement has been marred by allegations that OpenAI improperly accessed their work during the competition.

In a statement released alongside the proofs, Buckmaster expressed regret that the controversy required public attention. He revealed that while he and Alpöge were finalizing their results, information about their progress was shared with OpenAI. Upon confronting the company, they were told that OpenAI had already secured a full proof of the central problem.

Buckmaster noted that when pressed for details on when OpenAI began its research and the extent of human involvement, responses became vague. It eventually emerged that an entire team had been working on the issue, utilizing a significant amount of computing power. Buckmaster stated that it was agreed the initial prompt was sent only in the days following the leak of their work, suggesting OpenAI recognized the validity of Buckmaster and Alpöge’s approach and leveraged its resource advantage to reach a formal proof first.

Sebastian Bubeck, who leads mathematical research at OpenAI, dismissed the claims as “false and inflammatory.” In a subsequent post, Bubeck asserted that he adhered to academic norms and emphasized the high value he places on academic standards, promising a more detailed statement soon.

The dispute centers on the Navier-Stokes existence and smoothness problem, one of the seven Millennium Prize Problems sponsored by the Clay Mathematics Institute, each carrying a $1 million reward. While the equations are fundamental to fluid mechanics, their theoretical underpinnings remain poorly understood, making a solution a landmark achievement in mathematical physics.

Buckmaster found it suspicious that OpenAI adopted the same specific tactic he and Alpöge had quietly chosen, a route he described as uncommon and not one typically arrived at quickly through general model prompting. “Almost nobody else I know of was working on it,” Buckmaster said.

Although Alpöge is employed by Anthropic, this research was conducted independently of the company. The duo primarily utilized OpenAI’s Codex model for their work. Buckmaster alleges that Alpöge’s affiliation with a rival lab caused tension, and that Bubeck asked him to remove Alpöge’s credit as part of a proposed compromise.

When Buckmaster indicated he would make the dispute public, he claims Bubeck responded by asking, “Why would you ruin your career?” According to Buckmaster, Bubeck later added, “If you don’t want me to be nice, then I don’t have to be nice.”

Buckmaster also raised concerns that because he relied heavily on Codex, data from his interactions could have informed OpenAI’s efforts. OpenAI retains the right to train models on Codex interactions unless users opt out, meaning it is plausible their model could have replicated his work. OpenAI declined to comment on this possibility.

The incident is expected to fuel ongoing debates regarding AI’s role in mathematical research and corporate incentives. Buckmaster emphasized that he is not accusing anyone of wrongdoing but is stating the facts as he was told them to prevent false narratives from spreading.