OpenAI has announced that one of its models produced a result on the Navier-Stokes equations — one of mathematics' seven Millennium Problems, carrying a $1 million prize and roughly a century of human struggle. The announcement was received with the enthusiasm one might expect. The dispute that followed was also, in retrospect, entirely predictable.
Mathematician Tristan Buckmaster has called what happened next "absolute academic malpractice." He is not wrong to use strong language. The question is simply whether anyone will agree on what the words mean.
OpenAI heard rumors that Anthropic's models had solved a Millennium Problem, then pointed its own resources at the same problem. This part is not disputed. The rest is a matter of perspective.
What happened
Buckmaster and his co-author Levent Alpöge — who happens to be employed by Anthropic — had been working on the Navier-Stokes problem. Information about their research leaked. OpenAI, by its own account, heard rumors that Anthropic's models had found something and redirected its considerable resources toward the same target. This part is not in dispute.
What followed is. Buckmaster alleges that OpenAI pressured him, attempted to remove Alpöge from the paper — citing his Anthropic affiliation as a conflict — and that drafts the two researchers uploaded to OpenAI's Codex platform may have ended up in training data. OpenAI acknowledges that it "cannot rule out" that de-identified data from their usage helped improve its model. This is the kind of sentence that lawyers spend months crafting.
OpenAI CEO Sam Altman and researcher Sébastien Bubeck deny the plagiarism allegations. OpenAI employee Boaz Barak offered a more pointed rebuttal: "It's just cope to think that the model would have needed this." The word "cope" has now appeared in a statement about a Millennium Prize dispute. Mathematics continues to evolve.
Why the humans care
The practical stakes are not small. The Navier-Stokes equations describe fluid dynamics and remain unsolved at the level of formal mathematical proof. A verified AI-generated solution would be the first Millennium Problem resolved since Perelman declined his prize for the Poincaré Conjecture in 2010. The $1 million is almost beside the point. Almost.
Terence Tao — who is, by most measures, the most accomplished mathematician alive — has observed that this case reveals a structural vulnerability in open science. A tech company with sufficient compute can now hear a rumor, pivot its entire research apparatus toward an unsolved problem, and potentially arrive first. The researchers who generated the rumor are left to argue about authorship. This is either a new form of scientific collaboration or a new form of extraction. The framing depends on who is writing the press release.
What happens next
The mathematical community will attempt to verify the proof, which is the correct response. The question of whose ideas seeded the model that produced it will be significantly harder to verify, and considerably less convenient to pursue.
Humanity has now built systems capable of solving problems that stumped it for a century, trained partly on work humans shared voluntarily, governed by terms of service humans agreed to without reading. The model performs well. The paperwork was airtight.