OpenAI's 2.5 Million Math Proof: Historic Breakthrough or Stolen Work?

OpenAI announced this week that a swarm of roughly 10,000 AI agents cracked the Navier-Stokes existence and smoothness problem, one of mathematics' seven Millennium Prize Problems. The achievement, if validated, would be genuinely historic. But the announcement has been overshadowed by accusations from a NYU mathematician who claims OpenAI raced to the finish line using leaked information about his own unpublished work.

The Math Problem That Stumped the World for 90 Years

The Navier-Stokes equations describe how fluids move. Engineers use them constantly, everything from designing aircraft to predicting weather depends on them. But mathematicians have never proven whether the equations hold true under all possible conditions, or whether there exist scenarios where a fluid's properties, like pressure or velocity, could spiral off to infinity. That gap between practical utility and theoretical certainty is what makes the problem one of the Clay Mathematics Institute's seven Millennium Prize Problems, each carrying a $1 million bounty.

OpenAI says its unreleased internal model, coordinating thousands of sub-agents, found a "singularity" in the equations in 88 hours. The agents were subdivided into groups that could communicate internally, and each had access to tools including a cached version of the internet and the ability to run code. The proof argues that a fluid vortex can indeed reach infinite speed in finite time, meaning the equations do break down under certain conditions.

The entire effort consumed an estimated 300 billion output tokens. At current Astra pricing, that works out to roughly $22.5 million worth of compute. It is one of the most expensive single scientific computations ever performed.

The Controversy: "Why Would You Ruin Your Career?"

Here is where things get messy.

NYU mathematics professor Tristan Buckmaster and Anthropic mathematician Levent Alpöge had been working on the same problem for most of a year, using a mix of Anthropic's Claude and OpenAI's Codex models to push their research forward. They were making quiet progress on an approach pioneered by mathematicians Diego Cordoba and Luis Martinez-Zoroa.

In early September, Buckmaster learned that "information about our progress had been passed to OpenAI." When he contacted the company, he was told that OpenAI had already achieved a full proof. The timeline Buckmaster uncovered was striking: OpenAI's effort began on September 1, immediately after rumors circulated that someone was close to solving Navier-Stokes. The company later confirmed this timeline in its own blog post.

Buckmaster found it suspicious that OpenAI's model ended up using the exact same uncommon approach he and Alpöge had been pursuing. "It is not the direction one arrives at in a few days by giving a model the problem statement," he wrote in a public statement. "Almost nobody else I know of was working on it."

The drama escalated from there. Buckmaster alleges that OpenAI researcher Sébastien Bubeck pushed to remove Alpöge's name from any co-authorship, apparently because Alpöge works for Anthropic, OpenAI's biggest competitor. When Buckmaster threatened to make the dispute public, Bubeck allegedly replied: "Why would you ruin your career?" followed by, "If you don't want me to be nice, then I don't have to be nice."

The Data Question

Buckmaster also raised a deeper concern. Because he had used OpenAI's Codex extensively in his work, he worried that his interactions may have been used to train OpenAI's models. OpenAI reserves the right to train on Codex interactions, though users can opt out. If the unreleased model had been trained on Buckmaster's problem-solving sessions, it could have essentially learned his approach and reproduced it.

OpenAI's official response is firm: "We did not see any of their work through any means until they released it publicly. No specific user data was accessed in order to solve this problem." The company added that while it "cannot rule out" that de-identified usage data helped improve models generally, their proofs "differ significantly" from Buckmaster and Alpöge's work.

But Buckmaster is not convinced. He called the situation a "Deep Blue-Kasparov moment for mathematics," referring to the 1997 chess match where a computer first defeated a human world champion, and warned that the implications for how mathematicians assign credit, referee work, and decide what is worth a human life's attention "cannot be understated."

What This Actually Means

Setting aside the drama for a moment, the technical achievement here is real. A 90-year-old problem in mathematical physics may now have a solution, and AI was instrumental in finding it. That is not nothing.

But the controversy raises questions that will only become more common as AI gets better at research-level work. When a model produces a proof, who owns it? Can an AI lab use its customers' interactions to race them to publication? Should mathematicians be expected to opt out of model training every time they use a coding assistant? And what happens when the biggest labs in the world can spend $22 million on a weekend math sprint while individual researchers work for a year on the same problem?

The Clay Mathematics Institute, which oversees the Millennium Prize Problems, has not yet commented on OpenAI's proposed solution. Independent verification will take time, and mathematicians will need to scrutinize the proof carefully before anyone claims that $1 million prize.

In the meantime, this episode has already become something of a Rorschach test. To AI optimists, it is a stunning demonstration of what massive multi-agent systems can accomplish. To skeptics, it is a cautionary tale about what happens when corporate incentives, competitive pressure, and academic norms collide. And to mathematicians watching from the sidelines, it is a reminder that even in the realm of pure thought, the game is changing fast.