OpenAI’s Navier–Stokes claim leaves mathematicians checking the work

What OpenAI claimed

A human who solved the problem would expect prize money and recognition from fellow mathematicians. OpenAI’s announcement produced something different: for some mathematicians, an “existential crisis” about what AI can do and accusations that the company had claimed credit for human work.

OpenAI cited sources in its paper. Critics nevertheless argue that the acknowledgement was insufficient, especially for several mathematicians believed to have worked close to a solution. Their concern is not merely about courtesy. If the contribution of those researchers disappears behind the company’s result, human mathematical ingenuity becomes difficult to see—and OpenAI receives the recognition.

Mathematician Tristan Buckmaster has also raised concerns that OpenAI’s team may have seen his work on the Navier–Stokes problem, which he conducted with OpenAI’s Codex model. OpenAI denies direct access to those materials, but has not ruled out that data obtained while Buckmaster used the company’s products “helped improve our model.”

The unresolved issue is therefore not only whether the announced solution is correct. It is also whether the path to that solution was properly understood, credited and compensated.

Mathematics still needs people

AI companies are happy to showcase their models’ mathematical abilities. But mathematics also exposes a weakness in that story: systems can produce results that are careless or difficult to understand, leaving people to determine whether the work is valid, useful and applicable.

That makes mathematical expertise part of the product, not just an external review layer. To establish whether OpenAI’s claimed solution contains new mathematical tools or ideas, researchers will need time and intellectual effort. A striking final proof is not the same thing as a contribution that the field can understand and use.

Film or a vaccine made with AI has an obvious practical purpose regardless of its quality. A mathematical result is different. Its value depends on clarity and understanding, and those qualities may only emerge after human mathematicians have examined the work.

The announcement could have been an opportunity for collaboration. Instead, it has become another source of alienation and dissatisfaction.

The result is not the whole contribution

Mathematicians are not uniformly hostile to AI. Many remain open to using it, including researchers who have signed open letters criticizing technology companies. Most acknowledge that the technology is powerful and particularly effective at mathematics.

The more interesting question is how mathematical work changes when AI becomes part of it. In the emerging model, people direct the system’s work and interpret its results. That leaves an awkward problem for any company presenting a machine-generated proof as the central achievement: should credit belong only to whoever—or whatever—produced the impressive final artifact?

Mathematician and Fields Medal winner Bill Thurston wrote in 2010 that the product of mathematics is clarity and understanding, not theorems by themselves. That distinction matters here. A model may generate a proof, but the field still needs people to explain what it means and whether it introduces anything genuinely new.

I think this is why the Navier–Stokes announcement has created more unease than celebration. The technical capability is difficult to dismiss, while the system around it still offers no convincing answer for how human work should be recognized or paid for.

A wider conflict over AI

Some AI critics continue to treat all machine-generated results as equally useless “garbage.” That view is unlikely to survive every domain. AI will not perform well on every task, but more people may find themselves in the position mathematicians now occupy: convinced by the technology’s effectiveness and deeply worried about how companies deploy it.

Mathematician Nestor Guillén recently wrote that this anxiety might disappear if technology could be separated from technology companies and its use made democratic. That is a much larger demand than better attribution in one mathematical paper. It points to a conflict between the usefulness of AI and the institutions controlling access to it.

What OpenAI’s announcement leaves quiet is the part mathematicians may care about most: who gets to define the contribution, verify the result and receive recognition for the work beneath the headline. Until those roles are visible, even a successful AI proof will look less like the end of a problem than the start of an argument.