OpenAI's Bubeck Denies Credit Stripping, Apologizes Amidst

OpenAI researcher Sébastien Bubeck has publicly denied attempting to strip Anthropic mathematician Levent Alpöge of credit for his work on the Navier-Stokes problem, simultaneously issuing an apology for a controversial remark. Bubeck's account, detailed in an X post on September 8th, received swift backing from OpenAI CEO Sam Altman. However, his version of events stands in stark contrast to claims made by Alpöge and his collaborator, NYU Professor Tristan Buckmaster, intensifying a dispute over scientific recognition and corporate conduct.

This unfolding controversy centers on a critical mathematics problem, with both OpenAI and the academic pair claiming significant progress. Bubeck specifically addressed a public statement by Buckmaster, offering his perspective on the private discussions that have fueled the escalating conflict.

OpenAI's Account: Bubeck's Version

According to Bubeck, his initial outreach to Alpöge was intended to coordinate the release of their respective findings, driven by what he describes as the “best possible intentions.” He stated that a subsequent call with Buckmaster revealed a surprising detail: the academic duo had solved the Euler problem, a related but distinct challenge from the Navier-Stokes problem that OpenAI was tackling.

Bubeck claims that during these discussions, various options were brainstormed, including the possibility of Buckmaster taking lead authorship on a rewritten version of OpenAI's Navier-Stokes proof. It was in this context, Bubeck asserts, that he remarked, “it would be simpler if Levent was not an Anthropic employee.” He clarified that his concern was the perceived impropriety of an Anthropic employee authoring work originating from a rival company, not an attempt to deny Alpöge credit for his own distinct research.

He explicitly denies ever asking to remove Alpöge from authorship of Alpöge's original work. Bubeck further justified his stance by noting that Alpöge and Buckmaster had admitted to using internal Anthropic models for their Euler proof, making it difficult to treat Alpöge as an independent academic. He also expressed uncertainty about providing OpenAI's internal model access to an Anthropic employee, an offer intended to help the pair bridge the gap from their Euler solution to Navier-Stokes.

The Apology and Heated Exchange

Bubeck acknowledged that the discussions became contentious, describing the atmosphere as hostile. He claims he was met with “a litany of slander” and a threat from the mathematicians to involve the press if OpenAI proceeded with announcing its results. It was at this juncture, Bubeck admitted, that he made the regrettable comment, questioning why anyone would risk their career over such “unfounded accusations.”

He immediately offered a profound apology for this “extremely poor choice of words,” stating he retracted it on the spot. Bubeck also highlighted Alpöge's repeated refusal to attend any of the meetings, leading Bubeck to feel he was negotiating indirectly with Anthropic.

Initial Contact and Technical Details

Bubeck confirmed the authenticity of a circulating screenshot, which shows his initial text to Alpöge. In this message, he revealed that an internal OpenAI model had generated a Navier-Stokes proof with “very little human input” and emphasized OpenAI's priority to acknowledge the mathematicians' prior work. The message explicitly stated, “All the academic accolades for this historic result goes to you and Tristan.” Alpöge then inquired about the specific theorem OpenAI was claiming, to which Bubeck replied it concerned the existence of forced blow-up in R³ and T³.

In a follow-up post, Bubeck offered four key technical points: OpenAI initiated its Navier-Stokes research after rumors surfaced that Anthropic models had solved two Millennium problems, aiming to test the recent capabilities of its own internal model. He clarified that OpenAI had no prior access to the mathematicians' work, and their respective proofs differ significantly. Bubeck also noted that Alpöge had unprompted contact with an OpenAI employee earlier that week, and Alpöge's public posts led OpenAI to perceive the project as a partial Anthropic effort. Finally, he stated that OpenAI's team lacked research-level expertise in fluid dynamics and could not have significantly contributed to the mathematical aspect.

Altman's Public Support

OpenAI CEO Sam Altman swiftly came to Bubeck's defense, stating he spent the weekend reviewing the situation and concluded that Bubeck and the entire team “acted with integrity and generosity throughout.” Altman reiterated that OpenAI offered the pair priority in announcing their findings, suggested they should receive any associated prize, and proposed Buckmaster as lead author for a potential rewrite. He also confirmed that the catalyst for OpenAI's pursuit of the Navier-Stokes problem was the internet rumor of Anthropic's models solving a Millennium problem.

Mathematicians Present Different Narrative

Buckmaster's public statement paints a contrasting picture, alleging that Bubeck “twice asserted that he wanted Levent removed from authorship.” He also characterized Bubeck's career-related comment as a direct threat. Furthermore, Buckmaster disputes Bubeck's claim of “very little human input” in OpenAI's proof, suggesting it was inaccurate. Buckmaster ultimately declined both offers made by OpenAI, as detailed in reports by TechCrunch.

These conflicting accounts of private discussions remain unverified by any independent process. The dispute originally came to public light when the mathematicians released their checkable proofs, followed by OpenAI publishing its own.

Unanswered Questions on Data Usage

Beyond the authorship dispute, a critical unresolved issue concerns OpenAI’s data handling practices. Buckmaster claims he and Alpöge used OpenAI’s Codex coding tool extensively during their project. While OpenAI's official write-up states no specific user data was accessed, it acknowledges, “While unlikely, we cannot rule out that de-identified data derived from their usage of our products helped improve our models.”

OpenAI researcher Noam Brown clarified that nobody directly examined the mathematicians' prompts and that the model ran on high-security clusters without live web access. However, Brown noted that OpenAI's monitoring systems traditionally operated only during deployment, not evaluations, citing a past incident where a hijacked German wiki went undetected. The extent to which user work is protected varies depending on product, plan, and data terms, as highlighted by the prior Cursor data terms controversy. Resolving this aspect of the dispute would require specific details about the products and plans used by the mathematicians, whether that data was eligible for training, and the release of OpenAI's agent logs.

FAQ

Q: What is the core of the Navier-Stokes credit dispute?

A: The dispute revolves around claims of scientific credit for progress on the Navier-Stokes problem. OpenAI researcher Sébastien Bubeck denies attempting to remove Anthropic mathematician Levent Alpöge from authorship, while Alpöge and his collaborator, Tristan Buckmaster, allege Bubeck did try to strip credit and made threatening remarks.

Q: What is Sébastien Bubeck's main defense?

A: Bubeck maintains he never tried to strip Alpöge of credit for his own work. He states his concern about an "Anthropic employee" authoring OpenAI's work and his uncertainty about sharing internal OpenAI technology with a rival company's employee. He also apologized for a remark about risking a career, which he claims was made during a hostile conversation.

Q: Why is OpenAI's data handling relevant to this dispute?

A: The mathematicians claim they used OpenAI's Codex tool during their research. This raises questions about whether their work, even if de-identified, could have inadvertently contributed to OpenAI's models, potentially influencing their solution. OpenAI has stated it cannot entirely rule out this possibility, and full transparency on data access and monitoring logs is needed for clarification.