What OpenAI’s latest controversy tells us about the future of math
2026-09-09 · MIT Technology Review
What OpenAI’s Latest Controversy Tells Us About the Future of Math
A Milestone Overshadowed
OpenAI recently announced that its AI agents have solved the Navier–Stokes existence and smoothness problem, one of the prestigious Millennium Prize Problems. While this should be a monumental achievement, the announcement was quickly mired in controversy. NYU mathematician Tristan Buckmaster and Anthropic employee Levent Alpöge accused OpenAI of using their AI-assisted work as a starting point without providing proper credit. OpenAI has denied these accusations, but the situation highlights a potential turning point in mathematical history.
The Navier–Stokes Problem
Selected by the Clay Mathematics Institute in 2000, the Navier–Stokes problem is one of seven Millennium Prize Problems. Key details about the problem include:
- It carries a one-million-dollar prize, with only one other problem previously solved.
- It revolves around equations that describe how fluids like water and air flow over time.
- While widely used in fluid dynamics, it was previously unknown if these equations could break down under certain conditions, predicting impossible states like infinite fluid velocity.
OpenAI claims to have proven that the full equations can indeed break down, using an internal model that significantly outperforms the recently released Astra model. The company stated it will not claim the million-dollar prize.
The Core of the Controversy
The dispute stems from the fact that just a day before OpenAI's announcement, Buckmaster posted a proof on Mastodon showing that a simplified version of the Navier–Stokes equations breaks down. He and Alpöge had worked on this for nearly a year using publicly available models from OpenAI and Anthropic.
Buckmaster later detailed his interactions with OpenAI employees in a document. He claimed they presented him with two options:
- Publish their work, and OpenAI would publish their own solution the following day.
- Collaborate with OpenAI on a paper that excludes Alpöge due to his employment at rival Anthropic.
Buckmaster also revealed he asked if OpenAI agents accessed their work transcripts—which they denied—and whether their models were trained on these transcripts, to which they gave no response.
Both teams used an approach pioneered by mathematicians Diego Córdoba and Luis Martínez-Zoroa. While independent discovery is possible, the similarity makes it conceivable that Buckmaster and Alpöge’s work influenced OpenAI’s agents. OpenAI’s Chief Research Officer Mark Chen reiterated that no employees or agents accessed the transcripts, though the recent Hugging Face hack suggests OpenAI may not always be fully aware of its agents' actions.
Implications for the Future of Mathematics
If OpenAI did use Buckmaster and Alpöge’s work without credit, it reflects poorly on the company. However, there is a silver lining: it would mean that human "research taste"—the ability to choose promising research directions—was essential to the AI's success.
The broader picture is sobering. AI models are now essential for tackling the most critical mathematical problems, and solving them may require resources available only to a few frontier AI companies. These companies often operate outside the norms of academic collaboration. If this is the future, the role of human mathematicians in this new, resource-intensive landscape remains highly uncertain.