# Science.org Scrutinizes OpenAI's Claim to Have Cracked Navier-Stokes Millennium Prize Problem > Source: > Published: 2026-11-15 21:16:52+00:00 # Science.org Scrutinizes OpenAI's Claim to Have Cracked Navier-Stokes Millennium Prize Problem In September 2026, OpenAI announced it had solved the Navier-Stokes fluid dynamics problem — one of mathematics' seven Millennium Prize challenges. The claim appeared in a flurry of releases that included progress on the Riemann hypothesis and, in late October, a proof of the 4D Yang-Mills existence problem, another Millennium Prize problem. *Science* magazine has now examined the landscape of AI-driven mathematical claims, with particular attention to what happens when a company asserts it has felled one of the field's biggest trees. The *Science* article devotes most of its attention to the Yang-Mills result, where OpenAI took the unusual step of appointing a validation committee chaired by Nobel laureate David Gross rather than releasing the 1,147-page proof itself. On 10 November, that committee released a validation of a formalized version of the proof in Lean, a computer language that mechanically checks mathematical arguments against axioms. As *Science* notes, "Although it doesn't mean the still-unreleased proof is correct, it's a sign of its logical consistency" [\[1\]](https://www.science.org/content/article/ai-rivals-tackle-one-physics-deepest-mathematical-challenges). But for Navier-Stokes, the scrutiny runs deeper — and comes from a different quarter. A paper by Alexander Bastounis, Fabian Circelli, and Anders C. Hansen, posted to mathandai.org on 11 September and later on arXiv, directly examines OpenAI's announced Navier-Stokes proof. Their focus is on autoformalisation: the process of translating a natural-language mathematical proof into a formal language like Lean so it can be mechanically verified [\[2\]](https://www.uio.no/forskning/arrangementer/andre/forskningsetikk/matematisk_forskning_og_forskerutdanning_ki_uio.pdf). The authors argue that this translation step "may offer no confidence in the original NL [natural language] argument." The problem of resolving ambiguities in mathematical natural language well enough to produce a semantically faithful translation sits at the top of the Solvability Complexity Index hierarchy — informally, harder than the Halting problem. In practice, they demonstrate mismatches between natural-language proofs and their Lean "verifications," including for OpenAI's announced Navier-Stokes result. "The formalised Lean proof does not correspond to the NL proof of blow-up of solutions to the Navier-Stokes equations," they write [\[2\]](https://www.uio.no/forskning/arrangementer/andre/forskningsetikk/matematisk_forskning_og_forskerutdanning_ki_uio.pdf). This distinction matters. A Lean verification confirms that the formalized statements are logically consistent with Lean's axioms. It does not confirm that the formalization accurately captures the original human-readable proof — or that the original proof is correct. The *Science* article echoes this caution: the Yang-Mills Lean validation is "a sign of its logical consistency," not a certificate of truth for the unreleased manuscript [\[1\]](https://www.science.org/content/article/ai-rivals-tackle-one-physics-deepest-mathematical-challenges). OpenAI's September announcement was followed in October by a dump of more than 700 preprints on GitHub, an event *Nature* reported left "mathematicians are up in arms" [\[2\]](https://www.uio.no/forskning/arrangementer/andre/forskningsetikk/matematisk_forskning_og_forskerutdanning_ki_uio.pdf). The Advisory Group on Mathematics and Artificial Intelligence, hosted at the Institute for Advanced Study, has called for the immediate release of the Yang-Mills proof. Martin Hairer, a mathematician at the Swiss Federal Institute of Technology Lausanne, told *Science* that the Yang-Mills problem's "mountain of calculations made the problem a natural candidate for AI" [\[1\]](https://www.science.org/content/article/ai-rivals-tackle-one-physics-deepest-mathematical-challenges). Navier-Stokes presents a similar mountain of calculations. But the Bastounis-Circelli-Hansen paper underscores a structural issue: if the only artifact is a Lean formalization produced by the same AI system, the verification provides no confidence in the original proof. Ethan Sussman, a mathematician at Northwestern University who collaborated with Anthropic on Yang-Mills, told *Science* that AI-generated proofs can be hundreds of pages long and filled with unfamiliar language. "The only way to read one of these AI-written manuscripts is with AI," he said [\[1\]](https://www.science.org/content/article/ai-rivals-tackle-one-physics-deepest-mathematical-challenges). OpenAI's choice of David Gross to lead the Yang-Mills validation committee also drew criticism. Hairer noted that Gross had previously dismissed efforts to put quantum field theories on rigorous mathematical foundations as "useless" and "ugly" [\[1\]](https://www.science.org/content/article/ai-rivals-tackle-one-physics-deepest-mathematical-challenges). As the *Science* article implies, the mathematical community expects a human-readable, peer-reviewed proof that withstands scrutiny over time — a standard not met by a press release or a Lean formalization of an unseen manuscript. What happens next is unclear. OpenAI could release the natural-language Navier-Stokes proof for independent examination. It could submit to a journal. Or the claim could join the long list of announced solutions that never materialized. Until then, the only verified result about the proof is that its formalized version in Lean has been shown not to correspond to the natural-language argument [\[2\]](https://www.uio.no/forskning/arrangementer/andre/forskningsetikk/matematisk_forskning_og_forskerutdanning_ki_uio.pdf). ## Sources 1. [AI rivals tackle one of physics' deepest mathematical ...](https://www.science.org/content/article/ai-rivals-tackle-one-physics-deepest-mathematical-challenges) (science.org) 2. [KI i forskerutdanningen](https://www.uio.no/forskning/arrangementer/andre/forskningsetikk/matematisk_forskning_og_forskerutdanning_ki_uio.pdf) (uio.no)