OpenAI says an internal AI system solved the Navier-Stokes existence and smoothness problem, one of seven Millennium Prize Problems set up by the Clay Mathematics Institute in the year 2000, each carrying a one million dollar reward. Only one of the seven, the Poincare conjecture, had ever been solved before, back in 2003. If the proof holds up, this is a genuine milestone for what AI can do in pure math.
But the announcement landed in the middle of a credit fight. Tristan Buckmaster, a math professor at NYU, and Levent Alpoge, who works at Anthropic, had spent close to a year chasing the same problem using publicly available AI tools. They posted a partial breakthrough the night before OpenAI revealed its own complete solution.
Buckmaster says OpenAI gave him an ultimatum: publish jointly on OpenAI's schedule, or drop Alpoge from the paper entirely because Alpoge works for Anthropic, OpenAI's main competitor. He also says he asked OpenAI staff directly whether its AI agents had seen his and Alpoge's private conversations with AI chatbots, and got no clear answer. OpenAI denies any of this happened and says its team was only chasing a rumor.
Nobody outside OpenAI has independently verified the proof yet, and the Clay Mathematics Institute still lists the problem as unsolved on its own site. OpenAI has also said it will not collect the one million dollar prize, which is its own kind of statement.
The part that should matter most to any business leader is the cost. OpenAI says it ran about 10,000 AI agents at the same time for roughly 88 hours, spending millions of dollars in computing power to get the answer. That is a spending level that almost no university, research lab, or ordinary company can match. When a well-funded lab can throw that much money at a single hard problem, the smaller player with real expertise but a fraction of the budget increasingly loses the race, even if they got there first in spirit.
There is a second lesson buried here that applies well beyond mathematics. Buckmaster and Alpoge did serious, original work using AI chatbots from two different companies, openly and honestly, and a rumor about their progress was apparently enough to point a much bigger, better funded competitor toward the same target. If you are running a project through an AI assistant and discussing it with colleagues, assume that word of what you are working on can travel, even if no actual data is copied. Confidentiality in the age of AI tools is not just about who has access to your files. It is also about who overhears the fact that you are close to something valuable.
Fields Medalist Terence Tao warned last week that solving math problems purely through AI, without showing the work, could hurt the field rather than help it, because the mistakes and dead ends along the way are often where real progress comes from. The same logic applies to any company: if AI lets you skip straight to an answer, you also skip the learning that used to come from getting there the hard way.