A dispute over credit for a possible solution to one of mathematics’ hardest problems erupted this week after an NYU professor accused OpenAI of exploiting his group’s progress.
Tristan Buckmaster and Levent Alpoge, a researcher at Anthropic, published three preprints on September 8 proving finite-time blowup for the 3D Euler, 2D Boussinesq and porous medium equations under smooth forcing. All three results carry formal proofs verified in the Lean assistant.
In a statement, Buckmaster said his team learned that details of their work had reached OpenAI. When they asked how long the company had been pursuing the problem, he said the answers turned evasive, and that it emerged OpenAI had started only recently after hearing about their approach. OpenAI told them it already held a full proof of finite-time blowup for the Navier-Stokes equations, one of the Clay Mathematics Institute’s Millennium Prize Problems.
Sebastian Bubeck, who leads OpenAI’s math research, called the claims false and inflammatory. He told reporters OpenAI began training a specialized model on August 28 after seeing rumors that Anthropic was making progress, then ran more than 1,000 agents for over 50 hours before reaching a Lean-formalized solution over the weekend. Research head Mark Chen said the run cost millions of dollars.
The Clay Institute still lists Navier-Stokes as unsolved, and OpenAI had published no proof or technical paper as of September 8, leaving the rival accounts unresolved.
