OpenAI's claimed Navier–Stokes breakthrough ignites scooping and data-use dispute with mathematicians

OpenAI says an internal model running ~10,000 agents for 88 hours produced a Navier–Stokes Millennium Problem result one day after mathematicians Buckmaster and Alpöge posted related work, triggering allegations of scooping, pressure, and possible training on…

OpenAI announced that one of its unreleased internal models — described by MIT Technology Review as outperforming the recently released Astra — took 88 hours running a swarm of roughly 10,000 AI agents to produce a Navier–Stokes result: The Verge characterizes it as a solution to the $1 million Clay Mathematics Institute Millennium Prize problem unsolved by humans for nearly 90 years, while MIT Technology Review describes it as a proof that the full existence-and-smoothness problem can break down. The announcement came one day after NYU's Tristan Buckmaster and Anthropic's Levent Alpöge posted a proof of a simplified version, following nearly a year of work with public OpenAI and Anthropic models; both proofs use the Córdoba–Martínez-Zoroa approach, raising questions of independent discovery versus influence. Sébastien Bubeck said OpenAI's team pursued the problem after hearing a rumor about the researchers' work. Buckmaster accused OpenAI of academic misconduct — alleging it pressured him after his research leaked, offered him either next-day publication or a collaboration excluding Alpöge over his Anthropic affiliation, sidelined Alpöge, and may have trained on drafts entered into OpenAI's systems, including his Codex sessions. OpenAI denies using their transcripts or specific user data, a denial chief research officer Mark Chen reiterated, though the company acknowledges it cannot rule out that de-identified data from the researchers' product usage helped improve its models. OpenAI published a blog post, CEO Sam Altman defended the team's conduct, and the company says it will not claim the $1 million Clay prize. Terence Tao warned the episode could discourage researchers from sharing work, reversing centuries of open science, and that rumors alone can trigger AI-powered races that flatten original research projects. The Verge's critics call the rushed, reportedly million-dollar effort a violation of academic norms around trust and openness, while MIT Technology Review argues frontier-scale compute may concentrate major mathematical breakthroughs at a few AI labs.

• OpenAI says an internal model (unreleased per The Verge; outperforming the recently released Astra per MIT Technology Review) ran ~10,000 agents for 88 hours to produce the Navier–Stokes result.
• OpenAI's announcement came one day after NYU's Tristan Buckmaster and Anthropic's Levent Alpöge posted a proof of a simplified version after nearly a year of work with public OpenAI and Anthropic models.
• Both proofs use the Córdoba–Martínez-Zoroa approach, raising questions of independent discovery versus influence; Sébastien Bubeck said the OpenAI team pursued the problem after hearing a rumor about the researchers' work.
• Buckmaster alleges OpenAI pressured him after his research leaked, offered next-day publication or a collaboration excluding Alpöge for his Anthropic affiliation, and may have trained on drafts he entered into OpenAI's systems, including…
• OpenAI denies using the researchers' transcripts or specific user data — chief research officer Mark Chen reiterated the denial — but acknowledges it cannot rule out that de-identified data from their product usage helped improve its…
• OpenAI published a blog post, Sam Altman defended the team's conduct, Alpöge says he would have welcomed lab cooperation, and OpenAI says it will not claim the $1 million Clay Mathematics Institute prize.
• Terence Tao warns the episode could discourage researchers from sharing work, reversing centuries of open science, and that rumors alone can trigger AI-powered races that flatten original research projects.
• The Navier–Stokes problem carries a $1 million Clay Mathematics Institute prize and had been unsolved by humans for nearly 90 years, per The Verge.