OpenAI Accused of Stealing Math Professor's Work
Left 50%
Center 0%
Right 50%
1 left · 0 center · 1 right
What happened
On Monday, NYU mathematician Tristan Buckmaster and Anthropic mathematician Levent Alpöge released papers reporting new results on fluid-dynamics equations related to the Euler problem, work they said they produced with assistance from OpenAI and Anthropic models. In an accompanying statement, Buckmaster wrote that after he told OpenAI of the planned posting, company scientist Sébastien Bubeck told him an internal model had produced a proof of the related Navier-Stokes Millennium Prize Problem — a $1 million Clay Mathematics Institute challenge on whether smooth three-dimensional fluid flows can develop singularities — proposed coordinated publication, and wanted Alpöge removed as an author because he works at Anthropic; Buckmaster also reported Bubeck asking, "Why would you want to ruin your career?" and saying, "If you don't want me to be nice, then I don't have to be nice," while adding that he had not seen OpenAI's proof, did not know whether the pair's data was used, and was "not accusing anyone of anything." On Tuesday, OpenAI announced that its Astra model had solved Navier-Stokes in 88 hours, congratulated Buckmaster and Alpöge, and said neither its researchers nor its AI agents saw the pair's work until public release and that "no specific user data was accessed," while stating it "cannot rule out" that de-identified usage data helped improve its models and that the proofs "differ significantly." Bubeck denied on X that he had "ever asked Levent to be removed from authorship of his own work," said he had tried to "coordinate our releases" after "viral Twitter rumors that Anthropic had resolved 2 Millennium problems," called circulating charges "false and inflammatory allegations," and later described the career remark as a "poor choice of words" meant to warn against "unfounded claims." OpenAI's terms of service state that it trains on user-submitted content unless users opt out.
Sourcing check · Fully on the record · 2 articles
2 of 2 articles cite a named source for the story. It was on the record from the first report. Tristan Buckmaster denied part of it on the record.
What these terms mean and how they're counted
- On the record · names a source 2 of 2
- On the record means the article cites a source by name for the story's main claim: a person, or an institution speaking officially, such as a company statement or a named agency's spokesperson. The opposite is an unnamed source. The count is how many of the counted articles cite at least one such source; the names are everyone named across them, each once however the articles spell it. Here: Tristan Buckmaster, OpenAI, Terence Tao and Sébastien Bubeck
- Primary evidence · cites a document or data 0 of 2
- Primary evidence means the article cites a record you could check yourself: a court filing or ruling, a dataset or report, a recording or transcript, an official statement or press release, market data. A quote is not a document. The count is how many articles cite one. Here: None cited.
- Credits another outlet 0 of 2
- The article says the story comes from another outlet's reporting (“first reported by Reuters”). Those articles relay the story; they are not independent confirmation. Here: Each article presents its own reporting.
- Unnamed sources 0 of 2
- The main claim is attributed to people who are not identified: “people familiar with the matter”, “officials said”. The count is how many articles do this.
- On-record denial 1
- A named person or institution says the claim, or part of it, is not true. It is on the record — someone signed their name to it — but it is not confirmation, so it never counts toward On the record. Criticizing or dissenting from a decision is not a denial. The count is how many people or institutions deny it; each is listed with the first article that carried the denial. Here: Tristan Buckmaster (New York Post, Sep 9, 17:07 UTC)
- How it developed
- When the first report landed and whether it named anyone; when the first named source and the first document appeared. Here: First report Sep 8, 19:00 UTC by Business Insider, naming its source.
- Not counted 0
- Opinion, analysis and video pieces, and articles too short to assess. Every count above is out of the counted articles. An article republished under a new headline is one article.
This is about who is cited, not whether the story is true. Each source below says what it names. How we count
Omitted — what each side leaves out
Unpacked
Right-leaning coverage treats the dispute as possible copying of the same prize solution, while leaving out OpenAI's claim that the two sides proved different things. Left-leaning coverage reports the company saying the proofs "differ significantly," including on the related Euler question the professors actually published. Without that rebuttal, a New York Post reader is left with a theft story rather than a fight over unlike results.
Right-leaning coverage leads with a lightning prize solution and a professor who says company scientists threatened him. Left-leaning coverage leads with whether customer chats can train a model that then beats the user. An 88-hour sprint opens the right-leaning account, a figure Business Insider never reports.
Did any independent mathematician check OpenAI's proof before the prize claim went out?
Bottom line
The New York Post can call it cribbed work only by skipping OpenAI's statement that even the Euler results are different.
The receipts — 6 facts one side's coverage lacks
- OpenAI said its proofs differ significantly from the pair's and that the Euler results proved are different (forced versus unforced)
- OpenAI said neither its in-house researchers nor its AI agents saw the pair's work until the public release
- OpenAI's terms say the company trains on data submitted by users unless they opt out
- Buckmaster alleged OpenAI scientist Sébastien Bubeck asked why he would want to ruin his career and said if he did not want Bubeck to be nice, Bubeck did not have to be
- Bubeck later said he meant to warn about career damage from unfounded claims and admitted a poor choice of words
- Buckmaster said Bubeck wanted Alpöge removed as an author because he works at Anthropic
The Left View
Business Insider presents the dispute as a broader warning about AI training data and scientific credit: that inputs typed into models can later help a system "beat you at your own game," a concern it says everyone should care about even outside mathematics. It stresses Buckmaster's own hedges — that he does not know whether the pair's data or chats were used and is "not accusing anyone of anything" — alongside OpenAI's statement that it "cannot rule out" de-identified usage data even as it said no specific user data was accessed and that the proofs "differ significantly." The outlet points to the company's terms, which allow training on user content unless customers opt out, and treats the episode as posing open questions about how researchers should share credit with AI labs and whether companies may use customer inputs to compete directly against scientists. It also records Bubeck's denial of the authorship-removal claim, his account of moving after "viral Twitter rumors" about Anthropic, and his description of circulating charges as "false and inflammatory allegations," while noting Scientific American called it the "biggest day in math in at least two decades."
The Right View
The New York Post casts the episode as OpenAI having possibly "cribbed" the professor's work and then having "threatened him" after trying to "strong-arm" him into a joint announcement. It leads with the reported Bubeck lines about ruining a career and not having "to be nice," and with Buckmaster's doubt that a result "not the direction one arrives at in a few days" appeared only "in the past few days" after the company learned of his research. The paper quotes his question, "Is it ethical to use customer's data to try to scoop their customer?" and notes Bubeck later called the career remark a "poor choice of words" while saying he meant to warn against "unfounded claims." Coverage underscores the spectacle of a nearly century-old problem claimed solved in 88 hours by OpenAI's Astra model.
Our Take (balanced)
Left-leaning coverage is strongest on the documented training regime: OpenAI's terms allow learning from user content without an opt-out, and the company itself said it "cannot rule out" that de-identified data from the pair's usage helped improve its models, even while asserting independent proofs that "differ significantly." Right-leaning coverage is strongest on the contested human conduct around credit and timing — the reported career warning, the disputed request involving Alpöge's authorship, and the compressed interval after OpenAI learned of the planned release — which Bubeck answered by denying the authorship request, citing rumor-driven haste to coordinate announcements, and conceding a "poor choice of words." The central unresolved tension is whether OpenAI's Navier-Stokes claim was independent of the pair's work and product usage, or whether knowledge of that work (and possibly their chats) shaped a rushed assertion of priority; public statements cannot settle that, because Buckmaster has not seen OpenAI's proof and OpenAI has not produced a paper trail that would resolve the data-access question.
2 sources
More stories
The week's bottom lines, in your inbox
One email a week: the five stories that mattered and what they actually mean. Free.