OpenAI Claims AI Solved Navier-Stokes Amid Academic Misconduct Allegations
Original: OpenAI Just Claimed a Huge Math Discovery. Some Academics Are Crying Foul
Why This Matters
The case raises urgent questions about AI lab ethics, research priority, and data privacy in competitive AI-driven science.
OpenAI announced on September 8, 2026 that its AI solved the 200-year-old Navier-Stokes equation—one of the Clay Millennium Prize problems worth $1 million—using over 1,000 agents across 50+ hours at a cost of millions of dollars. Mathematician Tristan Buckmaster has alleged OpenAI acted improperly after learning of his and Anthropic researcher Levent Alpöge's prior progress.
OpenAI announced it has produced an AI-generated proof of the Navier-Stokes equation, a landmark unsolved problem in mathematics listed among the Clay Millennium Prize problems, each worth $1 million. Sebastien Bubeck, mathematician and AI researcher at OpenAI, stated the company began training a new math-focused AI model on August 28, and after hearing rumors of Anthropic making progress on the same problem, redirected significant resources toward it. More than 1,000 AI agents worked on the problem for over 50 hours before arriving at a solution, which was then formalized using Lean, a mathematical proof verification language. Mark Chen, OpenAI's head of research, confirmed the compute cost ran 'in the millions of dollars.'
However, the announcement has been overshadowed by allegations from NYU mathematician Tristan Buckmaster. On Monday, Buckmaster and Anthropic researcher Levent Alpöge published documents detailing key advances on the Navier-Stokes problem, claiming they used AI tools including Claude and Codex. Buckmaster separately alleged that OpenAI became aware of their work last week and began dedicating significant resources to the problem. He further claimed that OpenAI offered him a proposal to publish a paper announcing the solution under an internal OpenAI model's name—without Alpöge's name included. Buckmaster also raised questions about whether OpenAI accessed their Codex usage logs, saying he was told the model 'didn't look up user data' but received no response regarding training data questions.
OpenAI executives, including Bubeck, flatly denied inspecting the pair's Codex prompts and stated they did not see any of Buckmaster and Alpöge's work until it was publicly released. Bubeck also stated OpenAI 'recognize[s] the priority' of the pair's work on unforced Euler equations.