OpenAI accused of stealing math proof edge from NYU team — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Tristan Buckmaster (NYU) and Levent Alpöge (Anthropic) announced three proofs Tuesday, including preliminary progress on the Navier–Stokes existence and smoothness problem — one of seven Clay Millennium Prize problems, each worth $1 million — using a mix of OpenAI's Codex and Anthropic's Claude models.
- OpenAI published a full Navier–Stokes proof produced by an unreleased next-generation model, a week-long effort that consumed 300 billion output tokens — roughly $22.5 million in compute at current Astra rates.
- Buckmaster alleges OpenAI's first prompt was sent only "in the past few days, after information about our work had reached OpenAI," and that the lab's team independently converged on the rare "smooth force" approach (Fefferman options c and d) that he and Alpöge had been pursuing — a route "almost nobody else I know of was working on."
- Sébastien Bubeck of OpenAI allegedly asked Buckmaster to strip Alpöge's name from the work as a compromise, then warned him: "Why would you ruin your career?" followed by "If you don't want me to be nice, then I don't have to be nice."
- Buckmaster flagged a regurgitation risk: OpenAI reserves the right to train on Codex interactions, and his heavy use of Codex could plausibly have informed a model that then reproduced elements of his approach.
- OpenAI's own post pushes back, stating researchers "did not see any of their work through any means" and that the two proofs "differ significantly" — though conceding it "cannot rule out" that de-identified usage data improved its models.
- Anthropic employee Alpöge was not acting on company behalf, leaving Buckmaster reliant on OpenAI's Codex and raising questions about whether rival-lab collaboration with OpenAI's tools is structurally precarious.
Why it matters: Both teams may have just cracked a $1 million Millennium Prize problem using AI — but the dispute exposes how a single lab's compute advantage and first-party training data on its own coding product can let it race ahead of publicly visible academic work, with Buckmaster now publicly alleging career-pressure tactics by an OpenAI VP-level researcher to suppress credit for a rival-lab collaborator.
Ask SkimNews
