OpenAI claims Millennium Prize math breakthrough amid research credit dispute
A solution to the Navier-Stokes equations produced by 10,000 AI agents is overshadowed by allegations that OpenAI used private researcher data.
OpenAI announced on September 8, 2026, that an internal AI model has solved the Navier-Stokes existence and smoothness problem, one of the most elusive challenges in mathematics. The breakthrough, which addresses a Millennium Prize Problem, has immediately sparked a fierce dispute over intellectual property and research credit.
To achieve the result, OpenAI deployed approximately 10,000 autonomous AI agents using an advanced internal model not available to the public. Starting around September 1, the agents worked for roughly 88 hours to produce a proof showing finite time singularity formation for 3D Navier-Stokes equations. The final result was formally verified using the Lean programming language, with GPT-6 Astra utilized during the verification phase.
The Credit Controversy
The announcement was quickly met with allegations from NYU mathematician Tristan Buckmaster and Anthropic researcher Levent Alpöge. The pair claim that OpenAI may have misappropriated their private research drafts, which were stored in Codex sessions while they were pursuing a similar mathematical route.
OpenAI has denied having direct access to the researchers' private data. However, the company admitted it "cannot rule out that de-identified data derived from their usage of our products helped improve our models." The dispute further intensified over authorship; OpenAI reportedly suggested that Buckmaster could be an author on the findings but proposed excluding Alpöge due to his employment at rival AI firm Anthropic.
Why It Matters
The Navier-Stokes equations, established as a Millennium Prize Problem by the Clay Mathematics Institute in 2000, carry a $1 million reward for a solution. Beyond the prize, the problem is fundamental to understanding how smooth 3D fluid equations can develop singularities.
More broadly, this incident exposes a critical trust gap in AI-assisted science. As researchers increasingly rely on proprietary frontier models to formalize proofs and explore conjectures, the boundary between "de-identified training data" and the theft of intellectual property has become blurred. Tristan Buckmaster noted that the significance of this shift regarding how credit is assigned and how research is refereed "cannot be understated."
What's Next
The mathematical community now awaits a full peer review of the Lean-verified proof to confirm if the singularity formation claim holds. Meanwhile, the industry is watching to see if this dispute leads to new legal or ethical standards regarding the use of researcher data in the training of proprietary models. Whether this represents a new era of AI-driven discovery or a cautionary tale of data misappropriation remains to be seen.