OpenAI says 10,000 AI agents solved a Millennium Prize math problem in 88 hours — and touched off a credit dispute
OpenAI announced this week that a swarm of 10,000 AI agents working in parallel produced a proposed solution to the Navier-Stokes singularity problem — one of the seven Clay Mathematics Institute Millennium Prize Problems, each worth $1 million — in roughly 88 hours. The company says it will not claim the prize even if the proof is confirmed.
What the Problem Actually Asks
The Navier-Stokes equations describe how fluids like water and air move. Mathematicians have spent over a century trying to determine whether a perfectly smooth fluid flow could theoretically break down into an infinite-speed singularity within a finite amount of time. OpenAI's proposed proof describes a fluid vortex that stretches and concentrates until its velocity grows without bound, while the system's total energy stays finite.
How OpenAI Did It
The effort used 10,000 autonomous agents passing messages to one another over roughly 88 hours starting late Saturday, consuming compute resources OpenAI described as costing “millions of dollars” — about 1,000 times more than the company's previous mathematical efforts. Chief research officer Mark Chen called it “a significant milestone for AI research” with implications for tackling other unsolved problems.
The Credit Dispute
Hours before OpenAI's announcement, mathematicians Tristan Buckmaster (NYU) and Levent Alpöge (an Anthropic researcher) independently published AI-assisted work on related equations, using a similar methodology the pair had developed over nearly a year. Buckmaster alleges OpenAI pursued the problem only after learning of his research, and says OpenAI pressured him to drop his co-author's name from public discussion of the episode. OpenAI denies direct access to the rival work, but acknowledged “while unlikely, we cannot rule out that de-identified data derived from their usage of our products helped improve our models.”
Verification Will Take Years
The Clay Mathematics Institute has not verified the proof. President Martin Bridson said the evaluation process is “deliberately unhurried” — a solution must survive peer review and roughly two years of acceptance by the mathematical community before the prize committee even considers it.
Why It Matters Beyond the Prize Money
Regardless of the eventual verdict, the episode highlights a new flashpoint for the AI industry: as labs train on aggregated queries from millions of users, a rival's unpublished research becomes difficult to fully firewall off, even unintentionally. That raises uncomfortable questions about what “solving” a problem with AI actually means when the training data itself may carry traces of competitors' work — a dynamic that echoes the broader industry fight over AI model distillation.
Originally reported by Phys.org. Read the original article for additional details.
View original source