The proof had 10,000 authors and the praise went to two
OpenAI recently announced that about 10,000 AI agents, directed by its mathematicians, had produced a proof on the Navier-Stokes problem, which asks whether the equations describing fluid flow can break down. It sits among the Millennium Prize Problems, each carrying a $1 million prize. By Quanta Magazine's account, the agents ran for 88 hours and exchanged almost 5 million messages.
Charles Fefferman of Princeton, who wrote the official problem description, said he was thrilled. He then said who the heroes were, and neither was a machine: Diego Córdoba and Luis Martínez-Zoroa.
What the heroes had actually done
Quanta reports that both AI-assisted teams built heavily on their earlier work. Martínez-Zoroa's 2021 doctoral dissertation pioneered analytic techniques that do not rely on computers at all, which made him and his adviser Córdoba, in Quanta's word, oddballs in the area. Córdoba's line about it: "I don't use AI: I have Luis." Of Tristan Buckmaster of NYU, who led one of the AI-assisted teams, Martínez-Zoroa said he would have liked to do it themselves, "but I'm very happy for him."
Credit is contested even with names on every paper
The teams disagree about who did what, and when. Science News reports that human verification of the proof is incomplete, and that OpenAI has said it will not claim the prize. Gregory Eyink of Johns Hopkins calls credit when AI is involved "the really serious ongoing problem." If that is the state of play in a field where every contribution is published and dated, consider an ordinary company, where an idea may have no paper at all.
How an idea loses its name at work
The path is mundane. Someone floats a suggestion in a meeting, or built a template two years ago. It goes into a prompt, and what returns is clean and well organized. It gets presented by whoever ran the tool, and nobody in the room knows where it started. No one decides to take the credit. The originator simply watches it arrive without them.
The damage compounds. People notice which ideas get airtime and which get absorbed, and they adjust what they offer. A team that credits AI-assisted work loosely teaches its best thinkers to hold early ideas back.
Put the provenance on the page
Reminders to be generous do not survive a deadline. A line on the document does. Add three short fields to anything AI-assisted that goes up the chain.
Started from: the person, and the setting where the idea first appeared.
Built with: which tool did what, in a sentence.
Checked by: the person who verified it before it left the team.
Then have whoever chairs the review read the first line aloud before discussion begins. Within a few weeks people start supplying it unprompted, and the people doing the original thinking hear their names in the room.
Agreeing these rules before the first dispute is a working-session job: Work out how your team credits AI-assisted work →
Keep reading: Using it and deferring to it produce the same document, not the same person · Two questions about any idea. You can only answer one of them yourself.
This is the thinking behind a Trusted AI Culture.
Sources: Quanta Magazine, "AI Has Solved One of Math's $1 Million Millennium Prize Problems," September 8, 2026. Science News, "AI may have solved one of math's biggest puzzles, raising controversy," September 9, 2026.
Written with AI in the loop: my idea, AI drafted and sharpened, my judgment on the way out. Every word is mine to stand behind. — Darren