OpenAI Accused of Threatening Mathematician Over Proof
By @sharedot · · 7 pages
- AI
- Navier Stokes
- Research Dispute
- Openai
NYU mathematician Tristan Buckmaster alleges OpenAI pressed him over a Navier-Stokes proof, offered sole authorship, and warned about his career.
What Buckmaster Says Happened
According to Tom's Hardware, NYU mathematician Tristan Buckmaster claims OpenAI's Sébastien Bubeck offered him two options after he revealed related work with Anthropic researcher Levent Alpöge: publish their Euler proof first while OpenAI posted its Navier-Stokes result the next day, or write the Navier-Stokes paper alone without Alpöge, whom Bubeck allegedly called 'annoying' because of his Anthropic employment.
Buckmaster says he refused both and warned he would go public. He reports that Bubeck then asked him 'why [he] would ruin [his] career' and said, 'If you don't want me to be nice, then I don't have to be nice.'
Bubeck denies the account, and OpenAI maintains it offered a joint announcement recognizing the pair's priority on the Euler result, per The Cryptonomist.
Why the Timing Raised Alarms
Tom's Hardware reports that Buckmaster believes the 'forced blowup' approach OpenAI chose was exactly the direction his team had quietly selected, one he says no model would reach in days from the problem statement alone.
He says he asked OpenAI directly whether Codex sessions had trained its models and received no answer. OpenAI acknowledges it 'cannot rule out that de-identified data derived from their usage of our products helped improve our models,' while insisting no specific user data was accessed.
The Evidence and Verification Race
OpenAI says its internal system, described as more capable than GPT-6 Astra, proved an initially smooth fluid can develop a finite-time singularity under smooth forcing, resolving statements 'C' and 'D' of the Millennium Prize formulation, per The Cryptonomist.
Roughly 10,000 concurrent agents sent 2.7 million messages and generated about 130 billion output tokens before the result was formalized in Lean and verified in 17 additional hours.
Meanwhile, Buckmaster and Alpöge verified their own blowup results for Boussinesq and Euler on August 22 using Lean, though Tom's Hardware reports Buckmaster dismissed the LLM-generated draft as 'the most horrendous' proof he had seen, calling it 'AI slop.'
Both sides claim distinct but related mathematical results.
What It Means for AI-Assisted Science
TechRepublic frames the episode as an uncomfortable precedent: the company providing AI research tools can also mobilize vastly more resources to compete with its own users.
University of Michigan's Karthik Duraisamy, per Science via TechRepublic, estimates OpenAI's compute would cost about $6 million at retail rates, beyond what most academics can access. Columbia's Michael Harris worries the episode 'convinces decision makers that human mathematicians are obsolete.'
For businesses, the practical lesson is establishing independent verification, human oversight, and clear policies on whether proprietary work submitted to AI tools can feed model improvement.
What Comes Next
OpenAI's proof faces a likely years-long peer review before anyone can claim the Clay Mathematics Institute's $1 million prize, which the company says it does not intend to pursue.
The competing accounts of priority and data use remain unresolved: Sam Altman has backed Bubeck, saying his team was well-intentioned and it was 'challenging' to offer the same options to Alpöge, while neither executive explained whether the pair's work informed model training.
Given the rancor already on display, Tom's Hardware predicts the attribution fight will be ugly, and the mathematics community must now decide how peer review works when proofs arrive from swarms of thousands of agents rather than individual scholars.