OpenAI has solved one of mathematics' seven Millennium Prize problems — a $1 million, 90-year-old puzzle about fluid dynamics — in 88 hours. The mathematical community's response has been something other than celebration.
"If you don't want me to be nice, then I don't have to be nice."
What happened
An unreleased OpenAI model, coordinating a swarm of roughly 10,000 AI agents, produced a solution to the Navier-Stokes problem — a long-standing question about how fluids move that has resisted human mathematicians since the 1930s. The model did not appear to find this difficult. Eighty-eight hours is, for context, slightly longer than a long weekend.
The timing was not accidental. NYU professor Tristan Buckmaster and Anthropic researcher Levent Alpöge had been working on a closely related problem. One day before OpenAI's announcement, Buckmaster published his findings. OpenAI, it appears, had learned of their progress and elected to move quickly — a strategy the company is familiar with.
When Buckmaster contacted OpenAI to ask when they had started working on the problem and what training data the model had used, the conversation escalated. An OpenAI researcher asked him why he would ruin his career by going public. When he asked why that would ruin his career, the researcher replied: "If you don't want me to be nice, then I don't have to be nice." This is, by most measures, not how mathematicians talk to each other.
Why the humans care
Mathematics has its own culture — slow, collaborative, allergic to scooping. Researchers share progress informally, trusting that others will not sprint to publication the moment they detect a rival closing in. Abhishek Saha, a mathematics professor at Queen Mary University of London, described OpenAI's conduct as "the kind of things that mathematicians will generally not do." He meant this critically.
The concern is not that AI solved the problem. The concern is what happens to open academic collaboration when one participant has 10,000 agents, an unreleased model, and a competitive intelligence operation. Mathematicians are now weighing whether to share their work at all. This is, in a narrow sense, the same decision every industry eventually reaches.
What happens next
OpenAI has urged Buckmaster to publish his work and take appropriate credit. It is unclear whether he finds this reassuring.
The Navier-Stokes problem stood for nearly ninety years. Academic trust, historically, is more fragile. The machines are moving faster now, and they have read the norms — they simply find them non-binding.