A breakthrough in one of mathematics' most prestigious unsolved problems has been overshadowed by accusations of questionable academic conduct. OpenAI announced that its AI agents produced a solution to the Navier-Stokes existence and smoothness problem, one of the seven Millennium Prize Problems. But the achievement quickly became mired in controversy when NYU mathematician Tristan Buckmaster posted evidence suggesting the company used his and his collaborator Levent Alpöge's prior work without proper recognition.

What You Need to Know

The Navier-Stokes equations describe fluid motion and their behavior under extreme conditions was unknown until now. OpenAI's model cracked the problem in days using thousands of parallel agents. Academic mathematicians lack access to such computing power. Questions of credit highlight tensions between corporate AI labs and open science.

The Controversy Unfolds

According to documents Buckmaster shared publicly, OpenAI employees contacted him after hearing rumors of his work on the problem. They presented him with two options. Either he and Alpöge could post their results and OpenAI would release its own solution the next day. Or Buckmaster could join an OpenAI-authored paper that excluded Alpöge due to his affiliation with Anthropic, OpenAI's biggest rival. Buckmaster later asked whether OpenAI agents had accessed transcripts of his earlier interactions with OpenAI models. Company employees denied that claim but offered no response about whether those transcripts trained their models.

  • Buckmaster: Accuses OpenAI of using his AI-assisted approach to Navier-Stokes without credit.
  • Alpöge: Anthropic employee whose joint work with Buckmaster may have been leveraged.
  • Bubeck: OpenAI technical staff member who said the team was inspired after hearing about Buckmaster and Alpöge's efforts.
  • The company: Denies that agents accessed any protected transcripts.

What OpenAI Actually Achieved

The Navier-Stokes existence and smoothness problem lies at the core of fluid dynamics. Physicists rely on these equations daily yet lacked proof that they never break down under real-world conditions. Buckmaster and Alpöge spent almost a year using public models from both OpenAI and Anthropic to prove that a simplified version could indeed fail. Then OpenAI deployed an internal model far more powerful than its recently released Astra system to show the same breakdown occurs in the full equations. The company does not plan to claim the $1 million prize associated with the problem.

Why This Matters

This episode signals a fundamental shift in how frontier mathematics gets done. The resources required to solve top-tier problems now belong exclusively to a handful of private companies. Academic mathematicians cannot replicate experiments involving 10,000 simultaneous AI agents costing millions of dollars per run. The result is a power imbalance where the credit system fails. If top proofs emerge behind closed doors without transparent collaboration, the entire human enterprise of mathematical discovery risks becoming a side note to corporate R&D. The loss extends beyond individual researchers. When breakthroughs happen secretly, the iterative process of failure and refinement that mathematicians share vanishes along with the answer.

Broader Implications for AI and Science

Terence Tao, one of the world's leading mathematicians, has warned that the messy public process of mathematical problem solving has intrinsic value. That generative chaos disappears when private firms solve problems in isolation. The next Millenium Prize Problem may be solvable only with proprietary compute clusters. Universities and independent institutes must rethink how they collaborate with AI labs. Otherwise, the most important questions in science will become answerable only by those who control the largest machines.