OpenAI's Breakthrough Navier-Stokes Claim Draws Mathematician Rejection

OpenAI published a paper on Sept. 8 claiming its research system solved a piece of the Navier-Stokes smoothness problem, a Millennium Prize question carrying a $1 million bounty from the Clay Mathematics Institute, and mathematicians immediately pushed back on how much credit the company deserves.

The paper, posted at openai.com, says the work took roughly 88 hours of computation, according to the BBC. The lab has not clarified publicly whether the system used was Astra or Bel, per discussion tracked on Reddit’s r/accelerate.

The dispute centers on Princeton mathematician Theodore Buckmaster, whose prior published work on Navier-Stokes singularities appears to underpin large parts of the claimed result.

Fortune reports the lab contacted Buckmaster and offered him credit for the discovery, conditional on terms he found objectionable, prompting accusations of intimidation.

The Substance Of The Mathematical Claim

Navier-Stokes describes fluid motion, and the smoothness problem asks whether solutions to the equations can always be extended without blowing up into infinite values. Quanta Magazine notes the claimed result addresses a narrower piece of the full problem rather than the complete Millennium Prize question, worth asking, then, what exactly the 88-hour computation was measuring against, and whether the benchmark reflects independent derivation or elaboration on Buckmaster’s existing framework.

Also Read: OpenAI Posts First AI Proof for $1 Million Navier-Stokes Problem

Terence Tao and other mathematicians cited by Fortune and Axios have questioned whether the framing overstates the system’s independent contribution versus building on Buckmaster’s existing proof techniques.

Why The Credit Fight Matters For AI In Research

The episode exposes a structural gap in how AI labs claim scientific priority. Wired and Axios both describe a pattern where systems trained partly on published literature generate results that closely track existing human work, then get marketed as autonomous breakthroughs.

Scientific American frames the deeper stakes as institutional, if AI-assisted proofs can shortcut peer review’s credit norms, mathematics faces a fight over authorship rules well before it settles whether the underlying math even holds.