OpenAI Says Its AI Solved a 90-Year-Old Math Mystery in 88 Hours. The Controversy It Sparked May Last Longer.

Key points

• OpenAI says an unreleased AI model solved the Navier-Stokes problem, a fluid-motion puzzle unsolved for nearly 90 years, in just 88 hours.
• The company used roughly 10,000 AI agents, software programs that can carry out complex multi-step tasks on their own, running in parallel to find the solution.
• A New York University professor says an OpenAI researcher threatened his career when he asked questions about how the company learned of his related work.
• OpenAI admits it cannot fully rule out that anonymised data from users' chat sessions indirectly shaped its models.
• Mathematicians warn the episode could make researchers less willing to share early-stage ideas, damaging a field that runs on open trust.

OpenAI announced Tuesday that one of its AI models cracked the Navier-Stokes problem, a century-old puzzle about how fluids move that carries a one-million-dollar prize for whoever solves it. The achievement is real. The circumstances around it, however, have shaken the mathematics community to its core.

What exactly did the AI do?

The solution came in 88 hours. OpenAI pointed a swarm of around 10,000 AI agents, each one a software program acting on its own to handle part of the problem, at the Navier-Stokes equations and let them run. The company called it a "milestone" and said it has no intention of collecting the prize money, claiming its only goal is to show what its models can do.

The Navier-Stokes equations describe how water, air, and other fluids flow. Physicists and engineers use them constantly, from designing aircraft to predicting weather. Proving they always behave predictably, rather than breaking down into chaos, has beaten every human mathematician who has tried for close to 90 years.

Why are mathematicians angry?

The trouble started one day before OpenAI's announcement, when New York University professor Tristan Buckmaster published findings on a closely related problem. His co-author was Levent Alpöge, a researcher at Anthropic, the AI company that is OpenAI's main competitor. Alpöge was working on this privately, not on behalf of Anthropic.

Buckmaster says he contacted OpenAI after learning the company had somehow heard about their work. The conversation turned hostile fast. When he said he would go public, an OpenAI researcher reportedly replied: "If you don't want me to be nice, then I don't have to be nice." OpenAI, Buckmaster claims, also pushed him to publish the work crediting its internal model and to drop Alpöge as co-author entirely.

OpenAI researcher Sébastien Bubeck has disputed parts of Buckmaster's account, denying he asked for Alpöge to be removed. OpenAI itself told The Verge it found rumours of progress on Millennium Prize problems "on Twitter" and simply decided to try. It says it only later realised the rumours pointed to Buckmaster and Alpöge.

Did OpenAI spy on researchers' chat logs?

No confirmed evidence of that exists. But the company cannot give a clean denial either. OpenAI stated clearly that "no specific user data was accessed in order to solve this problem." At the same time, it admitted it "cannot rule out that de-identified data," meaning anonymised records stripped of names, "derived from their usage of our products helped improve our models."

Buckmaster had used OpenAI's Codex tool, a coding assistant, while working on the problem. He asked OpenAI directly whether his Codex sessions were accessed. He says the company became evasive.

For the wider maths community, even the possibility is alarming. Carnegie Mellon professor Jeremy Avigad put it plainly: "The thought that AI systems might steal ideas from our queries is chilling." Mathematicians are used to sketching half-formed ideas with colleagues, safe in the knowledge nobody will race them to a result. That informal trust, as Avigad warned, may now be harder to take for granted.

What does this mean for researchers going forward?

For most working mathematicians, day-to-day life will probably continue much as before. Only a handful of problems carry the fame and prize money to attract this kind of resource-intensive sprint. But the episode has exposed something uncomfortable: any company with enough computing power can now aim a swarm of AI agents at a problem the moment it hears a rival is close, at a cost of millions of dollars, and potentially get there first.

Mathematics professor Abhishek Saha at Queen Mary University of London said OpenAI engaged in "the kind of things that mathematicians will generally not do." University of South Carolina professor Matthew Ballard was blunter: the field "depends heavily on an informal norm of trust," he said, and that norm just took a serious hit.