ColdFusion: OpenAI’s Navier-Stokes Claim Sparks a Math Battle

SPEAKER_1 (0:01)

Nuance is rare online, but this story demands it. We need to talk about a week where OpenAI made a startling claim about solving one of mathematics' biggest problems, and everything that followed was messy, controversial, and deeply concerning. You're talking about the Navier-Stokes Millennium Prize problem, right? One of the hardest challenges in mathematics? Exactly. OpenAI claimed it had more than 300 examples of major progress on unsolved mathematics problems, with some reportedly solved outright. If independently verified, this could mark one of the greatest moments in mathematical history.

But here's where it gets complicated. The reaction isn't simply pro-AI or anti-AI.

Two things can be true at once: AI can be genuinely useful in coding, mathematics, and medicine, while also raising serious concerns about reckless leadership and irresponsible financial bets at these frontier labs.

The biggest flashpoint is definitely Navier-Stokes. On September 8th, OpenAI announced it had solved the problem, which asks whether fluid equations always behave normally or can reach impossible states, like infinite speed and energy.

This problem carries a one million dollar prize and had resisted the world's best minds for ninety years. But the celebration didn't last long. The controversy erupted over who actually did the work. NYU mathematician Tristan Buckmaster and his collaborator Levent Alpagay had been using LLMs like Codex and Astra in their research, storing everything in chat logs.

Right, and by late August, they say they had built the road toward solving the problem. Then Buckmaster spoke with OpenAI's Sebastian Bubeck and heard that an internal model had produced a 100-page proof using the exact niche method Buckmaster had been developing in secret. Buckmaster called that coincidence a "bright red flag." And it gets worse. OpenAI allegedly suggested the pair could publish the result and claim the prize, so long as Alpagay, who works at Anthropic, was removed from the credits. Buckmaster refused and says he felt pressured to stay quiet. OpenAI denied the accusations and called them "false and inflammatory," but the shifting explanation around data access only deepened suspicion. This is about more than one proof, though. It's about whether researchers can trust that unpublished work entered into AI tools won't be used to train or guide competing systems. Exactly. The Clay Mathematics Institute will review the proof carefully, but the damage may already be visible. Many academics now fear their work could be "scooped" or reduced to "benchmark fodder" in the race between powerful AI companies and the people trying to understand them.