On 8 September 2026 OpenAI said an internal model had solved the Navier-Stokes problem, one of the seven million-dollar Millennium Prize Problems. OpenAI's account: a rumour on 1 September, about a hundred AI agents on the simpler Euler equations, then some ten thousand on the prize problem, and a solution on 5 September, about 88 hours after the first agents started. The claim: a model of a fluid at rest whose core shrinks while its predicted speed grows without bound before a fixed time. No real fluid can do that, so the equations would fail. A computer has checked the proof. The Clay Mathematics Institute has not accepted it. Hours earlier, Tristan Buckmaster of New York University and Levent Alpöge of Anthropic published a related breakdown for the frictionless equations, after a year of private work with AI tools, including OpenAI's. Buckmaster's account of two calls on 6 September: two proposals, one of which would have had him write up OpenAI's result alone, without Alpöge. He refused both. OpenAI says it offered a joint release and to recognise their priority. Its researcher Sébastien Bubeck says he never asked for Alpöge's removal from his own work. On 10 September Buckmaster called that account false. Both proofs build on the work of Diego Córdoba and Luis Martínez-Zoroa. One question stays open. Buckmaster asked whether the model had trained on the drafts the pair typed into OpenAI's tools. OpenAI's Mark Chen says no human or agent looked at user data for this. OpenAI's post says it cannot rule out an indirect contribution through de-identified data. Whether their drafts helped is unknown. The object in the film is an illustration, not a measurement. Alpöge works at Anthropic, whose Claude models helped make this film. Sources: OpenAI post and paper, 8 Sept 2026; T. Buckmaster, statement, Euler paper and posts, 8 and 10 Sept 2026; Clay Mathematics Institute; S. Bubeck and M. Chen on X, 8 Sept 2026. #openai #mathematics #ai #navierstokes #news - @pollar.news