Core Event: OpenAI Claims Solution to Millennial Math Problem
On September 8, 2026, OpenAI announced its agents solved the Navier–Stokes existence and smoothness problem—one of seven Millennium Prize Problems selected by the Clay Mathematics Institute in 2000. Each correct solution qualifies for a one-million-dollar prize; prior to this, only one such problem had been resolved.
Key facts:
- Problem：Navier–Stokes existence and smoothness (fluid dynamics equations under conditions that may cause breakdown eg infinite velocity)
- Claimed solution：Proof that the full Navier–Stokes equations can break down under certain conditions
- Tool used：An internal model dramatically outperforming last week’s Astra model
- Resource scale：Approximately 10,000 agents running concurrently, millions of dollars
- Prize stance：OpenAI stated it will not claim the $1M prize
- Announcement date：September 8, 2026
Research Controversy: Attribution and the Boundary of Inspiration
Controversy centers on whether OpenAI relied on work by Tristan Buckmaster (NYU) and Levent Alpöge (Anthropic), who used publicly available models—including from both OpenAI and Anthropic—to cooperate for nearly a year. They recently posted a proof for a simplified version of Navier–Stokes breakdown on Mastodon.
According to Buckmaster’s public document, after he contacted OpenAI employees about rumors of their work, he was offered two options:
- He and Alpöge publish first; OpenAI publishes its full solution the following day;
- Buckmaster alone coauthor a paper with OpenAI, excluding Alpöge due to his Anthropic affiliation—the company’s largest competitor.
Buckmaster also asked whether agents accessed transcripts of his and Alpöge’s work; OpenAI denied such access occurred. When asked whether models were trained on those transcripts, OpenAI declined to respond.
Academic Reaction: Resource Gaps and Paradigm Disruption
Javier Gómez-Serrano, mathematics professor at Brown University, noted both teams use the Córdoba–Martínez-Zoroa approach, long considered promising yet one of several viable paths. Thus independent discovery remains possible—but not highly probable.
Notable contrast：Buckmaster and Alpöge’s nearly year-long collaboration produced only a simplified-result proof; OpenAI completed the full proof in days—a orders-of-magnitude efficiency gap between human and AI-assisted effort.
UCLA mathematician Terence Tao warned on Mastodon that “premature solving by purely AI-powered methods—especially without full transparency—can contaminate the research process and become a net negative for mathematics.” He emphasized that mathematics progresses through human missteps, wrong turns, and incomplete results, which collectively spur new theoretical tools.
Industry Implications: Reshaping Technical Barriers and Collaboration Norms
The case reveals deep tensions in AI-driven research:
- Resource imbalance：10,000-agent runs are financially inaccessible to nearly all academic labs;
- Cultural clash：Traditional math advances via open discussion and preprint sharing, while corporate AI labs favor closed development;
- Human “taste” dependency：If OpenAI agents followed the Córdoba–Martínez-Zoroa path because Buckmaster and Alpöge did first, human research intuition remains essential.
Practical Advice
Suitable for：Cross-disciplinary researchers in math and AI, academic leadership, and science-ethics policymakers.
Recommendations:
- if you support open science, engage in drafting attribution and provenance standards for AI-assisted proofs;
- if your institution plans to use AI agents for mathematical discovery, establish clear “inspiration traceability” policies in advance;
- avoid overreliance on single proprietary models—maintain investment in open, reproducible methods.
In Closing
Regardless of the truth, OpenAI’s announcement marks a pivotal moment: AI is now demonstrably engaged in core mathematical frontiers. Human mathematicians are shifting from problem-posed者 to potential “implicit inputs” training AI systems. Ensuring technical breakthroughs do not come at the cost of academic integrity is the next challenge the field must confront.
Note: This article is based on MIT Technology Review’s September 8, 2026 report. OpenAI spokespeople declined further comment.