OpenAI faces authorship pressure allegations in Navier–Stokes dispute

An OpenAI researcher is accused of pressuring an outside mathematician to drop an Anthropic-linked co-author, after a dispute over work connected to the Navier–Stokes existence and smoothness problem. TechCrunch reports that NYU professor Tristan Buckmaster, who worked with Anthropic mathematician Levent Alpöge, says OpenAI told them it had already produced a full proof after learning details of their approach. The Decoder adds that OpenAI described an internal, roughly 100-page proof involving Navier–Stokes with forcing during calls that Buckmaster says were pushed forward on short notice.

The argument sits at the fault line between two kinds of AI-era research: private labs with large compute budgets and academics using commercial tools that may retain rights to user interactions. Buckmaster and Alpöge say they used a mix of models, relying primarily on OpenAI’s Codex, and that they uploaded drafts into Codex sessions throughout the project. That matters because Buckmaster’s concern is not only priority in a “career-making” result, but whether the boundary between customer usage and internal research is legible enough for outsiders to trust the race.

According to TechCrunch, Buckmaster says OpenAI’s answers about timing and human input were evasive, and that the lab’s approach appeared to mirror an unusual route he says few mathematicians were pursuing. The Decoder reports that when Buckmaster asked whether OpenAI’s model had been trained on or had access to those Codex sessions, he was told the model did not look up user data, but he says he did not get a clear answer about training. In the same account, OpenAI’s Sébastien Bubeck denies wrongdoing, calling the allegations “false and inflammatory,” and says he will provide a fuller statement.

The personal stakes are sharpened by institutional ones. The Navier–Stokes problem is one of the Clay Mathematics Institute’s Millennium Prize problems, each carrying a $1 million award for a validated solution, and any claim of a proof must survive formal scrutiny. But scrutiny is harder when a result is said to have been generated with “massive compute resources” inside a private company, while outsiders are asked to accept timelines and process descriptions that are not independently auditable.

Buckmaster also alleges that OpenAI pressed him to remove Alpöge from authorship because of his employment at Anthropic, a point both TechCrunch and The Decoder report as central to the dispute. In Buckmaster’s telling, the conversation included warnings about reputational consequences if he went public; OpenAI has not, in the cited reporting, released a detailed account addressing the specific claims about authorship pressure and the provenance of the approach.

For now, the only concrete artifact described publicly is a claimed proof that has not been independently verified, and a set of phone-call recollections that neither side can substantiate with a paper trail. The Clay prize still requires a solution that can be checked by mathematicians who do not work for OpenAI.