What We Know About OpenAI's Disputed Navier-Stokes Math Claim

OpenAI amended its announcement claiming a solution to the Navier-Stokes existence and smoothness problem — a Clay Mathematics Institute problem carrying a $1 million bounty — after NYU mathematician Tristan Buckmaster said his own Codex prompts had informed the work, WIRED reported. The company deployed tens of thousands of agents to reach the result, according to Buckmaster, but he says it did so only after learning the equation was close to being solved.

What did Buckmaster accuse OpenAI of doing?

Buckmaster says he and Anthropic researcher Levent Alpöge had been using OpenAI's coding agent Codex and Anthropic's Claude to work on Navier-Stokes before OpenAI's September 8, 2026, announcement that it had solved the problem, WIRED reported. Buckmaster's contention is that OpenAI's system built on the trail of his prompts without fully crediting the underlying human work — a practice he calls "irresponsible" and "childish," particularly given the timing ahead of major AI company IPOs, per WIRED.

How did OpenAI respond to the dispute?

OpenAI investigated the claim and amended its Navier-Stokes announcement to state it "confirmed that Buckmaster's Codex prompts over the two months preceding this announcement and paper on September 8, 2026, could not have influenced the system in any way, including through training," according to the amended release cited by WIRED. The company pointed WIRED to that announcement in response to an inquiry. Buckmaster told WIRED that showing AI was pushing the boundaries of mathematics was "more important than the result" — even as he maintains the crediting process fell short.

Is this the only dispute over an OpenAI math claim?

No. German mathematician Andreas Thom raised a similar objection after OpenAI said in August that its Astra model had proved a long-standing geometric group theory problem Thom had spent two decades developing techniques for, WIRED reported. Thom says he emailed OpenAI researchers Mark Sellke and Sébastien Bubeck to point out that the firm's assertion that "no progress" had been made on the problem in the prior decade ignored a 2019 paper of his and other mathematicians' work. OpenAI amended that press release as well. Thom told WIRED he and a colleague had been using ChatGPT on the problem in the months before the result, and that when he asked whether their interactions had fed into training data, Sellke told him: "That did not happen."

Why do mathematicians keep using the same tools they distrust?

Despite the dispute, Buckmaster has continued relying on Codex — using it in the days since going public with his claims to tidy up research papers and to help him understand the logical steps OpenAI's agents might have taken to move from his earlier work toward the final proof, he told WIRED. "Even if you don't agree with any of this, you're kind of stuck. With AI being so useful, it's hard to completely prevent oneself from using it," Buckmaster said, adding that the companies building these systems "have a monopoly, and there is not much choice," per WIRED. Thom, for his part, told WIRED he has largely moved past the training-data question: "I set it aside. I'm not so much interested" in pursuing it further.

What does this mean for how math research gets credited going forward?

The episodes point to an unresolved tension in a field where AI labs are racing to claim solutions to problems that carry both prestige and, in Navier-Stokes' case, a formal cash prize. Mathematicians who supply the intermediate work — prompts, partial proofs, published papers — have limited means to verify whether that material shaped a model's output, and company statements that data "could not have influenced the system" or that interactions "did not" feed training rest on the companies' own internal reviews rather than independent audits, based on WIRED's reporting. Both OpenAI disputes were resolved through amended press releases rather than through a neutral third-party process, according to WIRED's account.

What to watch

• Whether the Clay Mathematics Institute or another outside body weighs in on crediting for the Navier-Stokes claim.
• Whether OpenAI or Anthropic publish methodology details showing how agents reached the disputed proofs.
• Whether other mathematicians surface similar concerns as AI labs continue publicizing solved problems ahead of anticipated public offerings.
• Whether Buckmaster or Thom pursue formal correction requests beyond the amended announcements.