Altman details OpenAI's side of a Navier-Stokes prize dispute

OpenAI CEO Sam Altman posted an account on X on September 8 defending the company's conduct in a dispute over credit for a proof touching the Navier-Stokes existence and smoothness problem, one of the seven Millennium Prize Problems. Altman said he "spent much of the weekend talking with the team who did this work" and said everyone involved, including an OpenAI researcher he named only as Seb, "acted with integrity and generosity throughout."

According to Altman, OpenAI's attempt began after the company heard rumors circulating online that Anthropic's models had already solved a millennium problem, and OpenAI wanted to find out whether its own model could do the same. OpenAI's team initially believed a rival group had solved the same problem OpenAI was working on. Wanting to coordinate rather than compete, Altman said, "we wanted to collaborate and do a joint release."

That changed once OpenAI learned the rival team had a result on the related Euler equations but not on Navier-Stokes itself. At that point, Altman said, OpenAI offered to let the other team publish first, suggested they should be the ones to receive credit for the prize, and floated the option of having a researcher he named as Tristan serve as lead author on a rewrite of OpenAI's own proof.

Altman said OpenAI felt it could not extend the same offer to a second researcher on the rival team, identified only as Levent, because he works at Anthropic, a direct competitor, and Levent "was not willing to talk or coordinate with us anyway." OpenAI said it held off publishing even though the other team was not communicating with it, and that it was met with what Altman called "unfounded accusations of plagiarism."

Altman said that now that both results are public, the two teams' mathematical approaches "appear to be different," and pointed to the underlying model's broader ability to "solve many, many other math problems" as the more significant story. He also confirmed the trigger for OpenAI's own effort was explicit: "It is true that we tried this because there were rumors on the internet last week that Anthropic's models had solved a millennium problem and we were curious if ours could do it too."

The episode is an early, concrete case of the credit and disclosure questions that arise when AI systems from competing labs converge on the same unpublished research result, particularly when a researcher affiliated with one lab is racing a rival lab's model toward the same proof. Neither side has published a full account reconciling every detail, and OpenAI's version, as related by Altman, is the only one so far confirmed for publication here.