OpenAI says an AI swarm made progress on the Navier-Stokes problem, but can't rule out that a mathematician's private notes helped. A red flag for private data, or coincidence?
OpenAI used a roughly 10,000-agent swarm on the Millennium Prize problem. NYU mathematician Tristan Buckmaster, who had used private Codex sessions on related work, noted the coincidence; OpenAI said it "cannot rule out" that de-identified data from his usage helped, while Buckmaster stressed he wasn't accusing anyone.
What happened
OpenAI announced its internal system, running roughly 10,000 concurrent agents exchanging 2.7 million messages, produced a proof addressing the Navier-Stokes existence and smoothness problem. NYU mathematician Tristan Buckmaster, who had stored private research notes in OpenAI's Codex product while working on related fluid-dynamics problems, flagged the methodological overlap. OpenAI's statement was a hedge, not a confession: 'While unlikely, we cannot rule out that de-identified data derived from their usage of our products helped improve our models.'
Why people disagree
Some say even a hedged admission that private research tools might double as training data for a company's own competitive claims is a serious transparency problem, regardless of intent. Others note Buckmaster himself explicitly said 'I do not know whether our data was used. I am not accusing anyone of anything' — and that OpenAI's phrasing is standard corporate caution, not an admission of wrongdoing.
What's still unclear
Without an independent audit of OpenAI's training data and logs, it may be impossible to prove or disprove whether Buckmaster's private notes materially contributed to the result.
Sources
More takes
House-hunters say AI-staged listing photos, which fill empty rooms with furniture that isn't there, are false advertising. Useful tool, or should they be banned?
Before today's US midterms, NPR got four of five AI chatbots to list candidates to vote for after a little pushing. Should chatbots ever suggest who to vote for?
