OpenAI Faces Allegations Over AI-Assisted Navier-Stokes Math Work

Mathematician Tristan Buckmaster says OpenAI pressured him after information about his AI-assisted work with Levent Alpöge on the Navier-Stokes equations allegedly reached the company. According to Buckmaster, OpenAI researcher Sébastien Bubeck wanted Alpöge removed from authorship because Alpöge works at Anthropic. Buckmaster says he refused and was then warned with remarks including, “Why would you ruin your career?” OpenAI denies the allegations.

Buckmaster and Alpöge used AI models while working on major progress related to the Navier-Stokes equations, which are listed among the Clay Millennium Problems. The Decoder reports that the pair used models including Anthropic’s Claude and OpenAI’s Codex running GPT-5.6 Sol. A recognized proof would be significant for mathematics and for AI-assisted research, but the article notes that some claimed results had not been published or formally verified at the time described.

Buckmaster says he and Alpöge uploaded all project drafts to OpenAI Codex sessions. When he asked whether OpenAI’s model had accessed those sessions, he says he was told the model did not look up user data; when he asked specifically about training, he says he received no answer. OpenAI later said no specific user data was accessed to solve the problem, while also saying it could not rule out that de-identified data derived from product usage helped improve its models.

OpenAI published its Navier-Stokes results and said the proof was produced by about 10,000 coordinated AI agents in 88 hours, using an internal model described as “significantly more capable than GPT-6 Astra.” The company said the proof was formalized in Lean and that compute alone cost “in the millions of dollars.” OpenAI said its researchers and agents did not see Buckmaster and Alpöge’s work until it was released publicly, and it said its proof differs significantly from their approach.

The dispute highlights practical questions for researchers using hosted AI systems: how drafts are stored, whether product usage can influence future models, and how priority should be handled when competing teams pursue similar breakthroughs. The safest takeaway is to review data-use settings, preserve independent records of work, and document timelines when using AI tools in high-stakes research. The case also shows how quickly AI-assisted math has become entangled with institutional competition between leading AI labs.