OpenAI Claims Major Math Breakthrough, but Accusations of Credit-Grabbing Emerge
OpenAI says its AI produced a solution related to the Navier-Stokes equations, a celebrated open problem in mathematics, but a rival mathematician alleges the company exploited private progress and tried to control credit.
OpenAI has announced that it produced an AI-generated solution to a long-standing mathematical problem connected to the Navier-Stokes equations, which describe how fluids such as air and water move. The company presented this as a demonstration of AI's potential in mathematics. However, the claim quickly became tangled in a dispute over attribution.
Tristan Buckmaster, a mathematician at New York University, and Levent Alpöge, a researcher at Anthropic, say they had been making progress on a related problem. Buckmaster alleges that OpenAI learned about their work and then rushed its own effort, while also attempting to influence who received credit. The Navier-Stokes equation is one of the Clay Mathematics Institute's Millennium Prize Problems, each carrying a $1 million reward.
According to Sebastien Bubeck, mathematician and AI researcher at OpenAI, the company began training a new AI model on August 28. He said that after reading rumors of Anthropic making progress on Navier-Stokes, OpenAI decided to invest more resources. More than 1,000 autonomous agents worked on the problem for over 50 hours before yielding a solution. Bubeck said he initially suspected an error, but by Sunday morning they had a final solution formalized in Lean, a programming language used for mathematical proofs.
OpenAI acknowledged that this problem demanded much more computing power than previous math tasks, with costs running into millions of dollars, according to Mark Chen, head of research.
On Monday, Buckmaster and Alpöge posted documents claiming key advances in a relevant area. They said they had used multiple AI models, including Claude and Codex. Buckmaster further alleged that he asked OpenAI whether the company had accessed their Codex logs. He said he was told the model did not look up user data, but questions about training went unanswered. He also described proposals from OpenAI, including one that would let him publish a paper saying the problem had been solved by an internal OpenAI model, but without Alpöge's name.
OpenAI denied using the pair's private work. Bubeck said the researchers and AI agents saw nothing until it was publicly released. He acknowledged the priority of Buckmaster and Alpöge's work on unforced Euler and congratulated them. Ven Chandrasekaran, a mathematician at OpenAI, said their solution was quite different in nature from OpenAI's output.
The dispute over credit may continue, and similar conflicts could become more common as AI plays a larger role in discovery. This is a developing story.
