Analysis
OpenAI's math-credit dispute, which Pulse first covered as it broke on September 8, escalated over the following day as OpenAI CEO Sam Altman put his own name behind the researcher at the center of the allegations. Altman said Sebastien Bubeck "acted with integrity and generosity throughout," according to Forbes -- a more full-throated defense than OpenAI's earlier, more hedged statement that it "cannot rule out" its models benefited from de-identified user data.
NYU mathematician Tristan Buckmaster also gave a fuller account of the private call at the center of his allegation. He says Bubeck presented him with two options once OpenAI's internal effort had produced its own proof: either OpenAI would publish the day after Buckmaster's team did, or Buckmaster could write up the result alone, with his collaborator Levent Alpoge -- who works at rival lab Anthropic -- left off the credit entirely. Buckmaster says he refused both terms and went public instead.
Terence Tao, not a party to the dispute but among the most-cited mathematicians alive, added a second, more systemic warning beyond his earlier comments. "We have now seen that even the rumor of someone working on a problem can trigger a massive amount of AI-powered effort to flatten it before the original research project has time to reach its full potential," Tao wrote, a claim that if accurate describes a new kind of competitive dynamic unique to labs with agent-swarm compute at their disposal -- one individual academics have no equivalent way to counter.
“NYU mathematician Tristan Buckmaster also gave a fuller account of the private call at the center of his allegation.”
Bubeck, for his part, held a press conference reiterating OpenAI's denial that its model had any access to Buckmaster and Alpoge's private work: "We did not use their prompts or proofs to prompt our models or direct our agents." That claim is, in principle, verifiable against internal model logs and timestamps -- evidence neither side has yet made public, unlike the competing accounts of the phone call itself.
None of this changes the underlying math. OpenAI's claimed proof of a Navier-Stokes singularity still has to clear formal peer review, and the Clay Mathematics Institute's own rules require a solution to be published and survive two years of scrutiny before the $1 million prize committee even convenes. What's changed since Tuesday is that the fight over who gets credit for getting there first now has OpenAI's CEO personally staked to one side of it.
For anyone running research on frontier-lab tools before publication, the practical lesson is already here regardless of how the credit dispute resolves: a product's default data-usage terms, more than its stated safety policy, determine whether unpublished work stays private while it's in progress.
