OpenAI announced that an unreleased AI model found an answer to the Navier-Stokes Millennium Prize problem, first reported by Singularityhub. The equations model fluids such as water and air, and some mathematicians have raised questions about credit and private work while prize conditions bar any award until at least two years after publication of a potential solution.
These equations underpin modern science and engineering. The question is whether fluid swirls can shift their energy into smaller, faster swirls, accelerating until the equations produce a blow-up in speed, also known as a singularity; OpenAI’s answer is yes under certain conditions.
The problem is one of seven Millennium Prize puzzles carrying million-dollar rewards, and only one had been solved before the announcement. OpenAI deployed a swarm of 10,000 AI agents running a new experimental model, using millions of dollars’ worth of computing power over a few days.
The model produced a solution in 88 hours, with verification taking another 17; the American Mathematical Society called the work the “final steps” of the solution process. The result built on work by several human mathematicians.
Tristan Buckmaster, a New York University mathematician, said he and Anthropic staffer Levent Alpöge had been using OpenAI’s publicly available models and pursuing a similar line of thinking. Buckmaster said he asked whether their data had been used by OpenAI’s model and received no answer; he also said OpenAI offered collaboration if Alpöge’s name was removed because of his Anthropic affiliation. OpenAI denies that claim.
German mathematician Andreas Thom suggested that OpenAI’s models were using unpublished human work and presenting it as AI-generated. Terence Tao has raised concerns about the indiscriminate use of powerful solution-extraction tools to achieve immediate short-term goals at the expense of broader understanding.
OpenAI insists that no specific user data was accessed by its researchers, but questions remain about whether the model was trained on Buckmaster and Alpöge’s private work. Companies and individuals around the world will now be re-examining how much they can trust OpenAI and other AI companies to handle their private and business data.
The model’s 88-hour result sits alongside a prize process that requires at least two years after publication of a potential solution before an award. For now, the Navier-Stokes problem is still officially unsolved.