AI Cracks Century-Old Math Problem, Stirs Credit Controversy

Imagine a challenge that has baffled the sharpest minds in mathematics for nearly a century, a deep mystery regarding how fluids move. We’re talking about the Navier-Stokes equations, a beast of a problem, one of the Clay Mathematics Institute’s coveted Millennium Prize Problems, complete with a cool $1 million reward for anyone who can finally crack it. Well, folks, it seems an AI might have just done the unthinkable.

On September 8, 2026, OpenAI made waves with a rather bold announcement: their artificial intelligence had, in fact, resolved the Navier-Stokes equations. A technical paper followed, detailing how an unreleased, internal AI model – described as "significantly more capable than GPT-6 Astra" – achieved this monumental feat. To give you a sense of the scale, we're talking about approximately 10,000 AI agents toiling away in parallel for 88 grueling hours, exchanging a staggering 2.7 million messages, and burning through an estimated 130 billion output tokens. The resulting proof, which posits that a smooth three-dimensional fluid initially at rest can develop a singularity, then underwent a rigorous 17-hour verification process using the Lean programming language to ensure its logical validity. Quite the computational marathon, isn't it?

Interestingly, despite this potential breakthrough, OpenAI has stated they don’t intend to claim the $1 million prize. This announcement came shortly after their public launch of GPT-6 Astra on September 3, 2026, a model itself trained on roughly 100,000 GPUs at their sprawling Stargate campus in Abilene, Texas. Greg Brockman, OpenAI President, even shared a tidbit that Astra marked the first time one of their models crossed a critical cybersecurity threshold. And in a moment of candid reflection, CEO Sam Altman quipped about the sheer cost, jokingly touching on the idea of AI being a bubble. You can almost feel the tension and excitement in those internal discussions!

But hold on a minute; no grand scientific discovery ever seems to be without its twists and turns, right? Just a day before OpenAI’s big reveal, on September 7, 2026, a significant credit dispute began bubbling to the surface. Two researchers, Levent Alpöge, a mathematician over at Anthropic, and Tristan Buckmaster, a mathematics professor from NYU, stepped forward. They claimed they’d been actively working on this very same problem, making substantial progress with the help of AI tools, including, ironically, OpenAI’s own Codex.

OpenAI, for their part, quickly countered. They explained that a rumor of parallel work reached them on September 1st, their proof was finalized by September 6th, and they subsequently reached out to Buckmaster and Alpöge, offering either a joint announcement or some form of credit arrangement. Sounds reasonable on the surface, doesn’t it? However, Buckmaster voiced his suspicion, suggesting OpenAI's project might have indeed drawn, perhaps indirectly, from his and Alpöge’s prior work, implying a lack of transparency from the AI giant. OpenAI, though, pushed back firmly, asserting their solution was "significantly" different and denying any access to "specific user data" for the problem's solution. Yet, and this is a crucial point, they also admitted, "While unlikely, we cannot rule out that de-identified data derived from their usage of our products helped improve our models." That subtle nuance really adds another layer to the whole controversy, leaving one to wonder about the blurred lines in data usage.

As you might expect, the academic world isn't rushing to judgment. Independent mathematicians are still carefully scrutinizing OpenAI's proof, and as of now, they haven't officially signed off on its validity. The authorship dispute, naturally, remains a very active conversation. It’s a bit like a high-stakes chess match where the next move is still pending. Ravi Vakil, a distinguished mathematics professor at Stanford University, perfectly encapsulated the challenge of the Navier-Stokes problem, calling it a "deep mystery" and a "huge thing that's out of reach." It makes you wonder, doesn't it, if AI has truly reached a point where such previously insurmountable hurdles can be overcome, and what that means for human ingenuity?