OpenAI's Navier-Stokes solution claim ignites backlash across mathematics

Over the past year, AI labs including OpenAI and Anthropic have announced a run of results on long-standing mathematical problems, in some cases going beyond what researchers expected from current systems. According to The Verge, which surveyed the mounting fallout, the reaction has been defined less by celebration than by conflict over how those results were obtained and released.

A Millennium Prize result announced under a cloud

According to The Verge, OpenAI used a blog post to announce that it had discovered a solution to the Navier-Stokes problem, which concerns the flow of liquids and gases and had stood open for roughly 90 years. It is one of seven Millennium Prize Problems, each carrying a $1 million reward, and The Verge notes the news was first reported by The New York Times and Wired.

The company said it reached the result with an internal model it considers more capable than its newly released GPT-6 Astra, run alongside 10,000 concurrent agents, with training of that model starting on August 28th. By The Verge's account, the achievement is genuine and a striking demonstration of how quickly AI is changing mathematics. But the path to it was troubled: the outlet reports that after hearing that other researchers were making progress, OpenAI appears to have redirected its considerable resources toward finishing first. The Verge says the ensuing controversy has surfaced allegations of scooping, spying and disregard for academic norms. Abhishek Saha, a mathematics professor at Queen Mary University of London, told the outlet that OpenAI engaged in the "kind of things that mathematicians will generally not do."

Disputes over where the training data came from

The Verge also reports a second front in the dispute: transparency around training data. Shortly after an earlier row over whether OpenAI's models benefited from unpublished academic work, mathematician Andreas Thom wrote in Mastodon posts that interactions he and his colleagues had with ChatGPT before OpenAI's announcements may have contributed to its success. One of the ten results in a recent OpenAI batch involved non-sofic groups, Thom's area of expertise, and The Verge notes OpenAI acknowledged that its result built heavily on prior work by Thom and mathematician Gábor Kun. Thom accused the company of unethical and dishonest behavior, according to the outlet.

An advisory panel, greeted with doubt

OpenAI's most recent attempt at repair is an independent panel of prominent mathematicians, announced without warning, tasked with advising OpenAI and other AI companies on their dealings with mathematical research and the wider community, including how new results are presented and released. The Verge reports that researchers see the group as a reasonable first step but are asking basic questions about what it will actually do, how much influence it will have and whether OpenAI will listen. There are also concerns about whether a small group of elite practitioners can speak for the field as a whole. Even one of the group's members described a messy and confusing process to The Verge.

The panel's opening assignment may be its hardest: helping coordinate the release of a large number of additional results that OpenAI says its unreleased model has produced. The Verge reports that the prospect is already causing dread among researchers worried about what a flood of machine-generated breakthroughs could do to their discipline.

A field reassessing itself

The tension extends beyond one company. James Maynard, a University of Oxford professor and Fields Medal winner, told The Verge he has spent much of the past year reflecting deeply on the future of a discipline that has traditionally moved slowly. The outlet frames the core unease this way: mathematicians feel OpenAI is doing mathematics for different reasons than they are. Researchers want to advance the field; the company, in their view, wants to win.

Why it matters

If AI systems can credibly attack problems at the Millennium Prize level, mathematics faces a structural shift. Norms around priority, credit and verification were built for human-paced work, not machine-paced output, and the current disputes show those norms straining. The training-data allegations are an early test of a broader question: whether researchers' unpublished work and casual chatbot conversations can legitimately fuel commercial models' results. How OpenAI handles its next wave of releases, and whether its advisory panel has real authority, will set a precedent for every other lab working at the frontier of research. The Verge's reporting suggests mathematicians are not rejecting AI assistance outright; the fight is over transparency, consent and respect for the conventions that hold the field together.