OpenAI Announces Navier‑Stokes Breakthrough, Sparks Math Community Debate

For the second time in history a Millennium Prize problem has been claimed solved, and this time the credit went to a machine‑learning lab rather than a lone human mind. OpenAI released a statement on Tuesday saying its in‑house model produced a proof that the famous Navier‑Stokes equations—used to describe fluid flow—are fundamentally flawed because they can develop infinite velocities, a scenario that nature simply won’t allow.

The company says the proof was formalised in the Lean theorem‑proving language, which many consider a strong safeguard against human error. In other words, the proof was checked by a computer, and the computer said “yes, it checks out.” That alone would be headline‑worthy, but the story got messy almost immediately.

Just a day before OpenAI’s announcement, mathematician Tristan Buckmaster posted on social media that he and his collaborator Levent Alpöge (who works at Anthropic) had managed to “blow up” the Euler equations—an achievement many in the field view as a crucial stepping stone toward a full Navier‑Stokes solution. Buckmaster alleged that OpenAI somehow caught wind of their work and then used the same “forcing” technique his team had pioneered.

OpenAI’s own mathematician, Sébastien Bubeck, pushed back hard. He told a press briefing that the AI’s internal model arrived at the Euler result independently, using a completely different line of reasoning. According to Bubeck, the AI only adopted a method similar to Buckmaster’s when tackling the broader Navier‑Stokes problem, and that work was completed over a weekend after the alleged leak. “We did not use their prompt or proofs to prompt our models,” he emphasized.

The controversy has set the math community buzzing. Diego Córdoba, one of the original developers of the “forcing” approach, described the news as “a bit of a shock.” Luis Martínez‑Zoroa, his co‑author, added that if the AI truly replicated their method, it would be an “unexpected surprise.” Meanwhile, University of Chicago professor Luis Silvestre remarked, “Yesterday and today are crazy days; we’re all trying to sort out what this means for mathematics.”

Regardless of who borrowed whose ideas, the proof’s reliance on Lean gives it a veneer of certainty that traditional, handwritten arguments often lack. Still, many mathematicians caution that AI‑generated proofs can be opaque—hard to trace, hard to understand, and sometimes hard to trust without deep human inspection.

OpenAI’s claim has reignited a broader conversation about the role of artificial intelligence in pure mathematics. If the proof holds up under peer review, it would mark the first time an AI has cracked a Millennium Prize problem outright, potentially reshaping how hard‑core theoretical work is approached. Until then, the math world will be watching, debating, and, inevitably, re‑checking every line of code.