Joint Statement by 25 Fields Medalists: A Week for OpenAI and the Math Community

Hotspot Tracking: Hotspot Release × Technical Judgment × Practical Advice. Author: 永亮

---

What Happened

On September 10, OpenAI released a technical report claiming that its AI system (approximately 10,000 agents collaborating, taking 88 hours, costing millions of dollars) made breakthrough progress on the Navier-Stokes equation problem. This is one of the Clay Mathematics Institute’s seven “Millennium Prize Problems”; a genuine solution carries a $1 million prize and requires rigorous verification by the academic community. Domestic media, when relaying the news, used the phrase “announced a solution”—but in the strict sense, a “solution” (proof certified by the Clay Institute) has not occurred, and a mathematician at New York University has publicly questioned the validity of the results.

The trouble is not limited to “whether it was truly solved.” Multiple mathematicians have accused OpenAI of scooping their ongoing Navier-Stokes research—a headline in The New York Times puts it well: “The Mathematician Crushed Between OpenAI and Anthropic.” One researcher stated that their exchanges with AI labs regarding the same problem ultimately became stepping stones for the other party’s results. The Economist used a sharp contrast for its headline: A year of work by two Spanish mathematicians versus OpenAI’s 88 hours and €15 million.

Thus came the joint statement on September 11. Terence Tao, Deng Yu, and 25 other Fields Medalists signed it. The core appeal is: The goals of AI companies and the mathematical community are diverging—the competitive pressure between laboratories is eroding the centuries-old open tradition of mathematics. Almost simultaneously, OpenAI withdrew its sponsorship of the Caltech Mathathon.

Why This Time Is Different

AI impacting mathematics is not news. In recent years, AI has solved Olympiad problems, assisted in proofs, and challenged Millennium problems; the mainstream attitude among mathematicians has been “cautious welcome.” The falling out this time is because the focal point of the conflict has shifted from “Can AI do it?” to “How do AI companies conduct themselves?”

The keyword in the accusations is scooping. The rule of mathematical research is: problems are open, the process is open, and credit belongs to the prover. The operating logic of AI labs is: competition first, release first, narrative first. When a player holding 10,000 agents and tens of millions of dollars in computing power enters the field, it can legally “finish” a route others are manually climbing, and then announce the summit via a press conference—authorship, acknowledgments, priority are all compressed into a single sentence in a press release. What angers Tao and others is precisely this: even if the technical result is real, the way it was obtained is breaking the game rules of this discipline.

There is an even colder question: If AI solves all math problems, what will mathematicians do? A Chinese Fields Medalist joked to the South China Morning Post: “Then I’ll go write romance novels.” Behind the joke lies a real existential anxiety—when “problem-solving,” the core function of mathematicians, is outsourced, where does the value anchor of this profession lie?

The Calm Other Half

The other side of the story must also be told. Regardless of whether OpenAI’s Navier-Stokes results meet the standard of a “solution,” its technical weight should not be dismissed by emotion: the work of 10,000 agents collaborating for 88 hours would have seemed like science fiction two years ago. AI-assisted proofs have indeed been helping mathematicians handle manual labor like verified proofs in recent years. Terence Tao himself has long been a user and advocate of AI tools; this statement is not anti-AI, but anti “unfair competition”—the spearhead of the statement points to the behavioral pattern of the laboratories, not the technology itself.

Similarly, the move to “withdraw Mathathon sponsorship” can be read as being “driven away” by mathematicians, or as OpenAI cutting losses to avoid escalating conflict. Currently, OpenAI has not formally responded to the statement and accusations; letting the dust settle for a while is safer than rushing to take sides.

The Real Takeaway

This conflict is actually a preview of the intellectual property wars in the AI era, except the battlefield has moved from code and content to humanity’s most abstract territory. It throws up three questions, each of which will appear repeatedly:

First, does compute priority equal academic priority? A team spends a year producing partial results, and an AI lab “verifies and expands” it in 88 hours—who owns the result? Current academic norms have no answer.

Second, when verification cannot keep up with release, how should claims be counted? There is a gap between “breakthrough progress” and “solving a Millennium problem” involving the Clay Institute’s certification process; this distance often disappears when media relay the news—OpenAI achieved a victory in the communicative sense, even if the mathematical victory is not yet in hand.

Third, can open tradition constrain closed laboratories? Mathematics runs on centuries of accumulated open tradition, while the competition among top AI labs is described as an “AI arms race.” There is currently no visible mechanism to expect the latter to follow the former’s rules, only open letter after open letter.

Tao and others issuing this statement is essentially using the century-old credit of the discipline to fight for rules for future researchers. It is unknown whether they will win, but the math community is perhaps the opponent that should least be underestimated—what this circle is best at is finding proofs in seemingly unsolvable problems.

References

• OpenAI Official Technical Report “On the Navier–Stokes Millennium Prize Problem” (2026-09-10)
• TechCrunch: OpenAI’s feud with mathematicians is only escalating (2026-09-11)
• The Economist: Top mathematicians are outraged by OpenAI’s methods; A year of work by two Spanish mathematicians versus 88 hours and €15 million
• The New York Times: The Mathematician Crushed Between OpenAI and Anthropic Over a Math Problem
• Chinese reports on the joint statement by Wall Street_cn, Sina Tech, and Zhixd Dong (2026-09-11/12); South China Morning Post interview with the Chinese Fields Medalist