OpenAI Solved a 200-Year Math Problem in 88 Hours. Who Gets Credit?

The claim arrived sideways

OpenAI reported that its unpublished AI model solved the Navier-Stokes existence and smoothness problem — one of the seven Millennium Prize Problems — in 88 hours. The story broke in Korean and Japanese outlets first. Western wire desks had not yet caught up.

That sequence matters. When AI accelerates discovery, it also accelerates citation. And the race to be credited — or to cite before being cited — starts the moment a result appears, not when it is verified.

What OpenAI actually claims

According to the report, the model known internally as something better than GPT-6 Astra began work on January 1 and reached a solution by January 5. Roughly 10,000 AI agents communicated with each other during the process, consuming millions of dollars in computing resources. The model proved that smooth initial conditions in the Navier-Stokes equations can maintain regularity over time — a question that has resisted proof since the equations were formulated in the 19th century.

Only one Millennium Problem has ever been solved: the Poincaré conjecture, proved by Grigori Perelman in 2002. The other six remain open. If OpenAI’s claim holds, the Navier-Stokes proof would be the second.

The speed is not the shock

What makes this moment different from AlphaGo is not velocity but domain. Go is a finite game with fixed rules. Mathematics is an open-ended practice — the rules evolve as the problems do. An AI that masters Go plays within a system. An AI that produces a mathematical proof enters territory where the system itself is under construction.

Professor Choi Kyu-don of UNIST in Ulsan said it plainly: this is more shocking than AlphaGo. “I think without AI this problem would not have been solved for 100, perhaps 200 years,” he said.

The Navier-Stokes equations describe how fluids move. They govern weather prediction, aircraft design, ocean currents. Two centuries of the world’s best mathematicians have worked on their existence and smoothness properties. No general proof has been found.

The credit question starts now

Here is what the academic community is already arguing about: if a machine produced the proof, who gets the citation?

Tristan Backmaster, a professor at NYU, said his own unpublished research may have been used by OpenAI’s model. He claimed he was actively working on the problem himself and that the AI’s result did not reflect the difficulty he experienced. “This is not a problem where you input it into a model and find a solution in a few days,” Backmaster said.

If Backmaster is right, the credit problem is not abstract — it is personal and immediate. An AI trained on unpublished preprints could effectively absorb a researcher’s life work and return a result faster than that researcher could publish it.

The question of who gets cited when no human wrote the proof is not theoretical. It is already happening.

What peer review looks like at AI speed

OpenAI’s model produced a result in 88 hours. Peer review will take months, possibly years. The gap between production and verification is the new friction in mathematical discovery.

Professor Lee Joon-sang of Yonsei University noted that the AI demonstrated a new research tool — one that can explore and formally examine mathematical proofs. But “the final validity and academic meaning require expert verification,” he said.

That verification step is where attribution lives. When a proof is reviewed, the reviewers decide not just whether it is correct but who deserves credit for the path that led to it. An AI that cannot be credited creates a vacuum — and vacuums get filled by whoever publishes first.

The institutional response

Some in the mathematics community are already speculating that awards like the Fields Medal — often called the Nobel Prize of mathematics — could become obsolete. If machines solve problems that took humans centuries, what institution awards the credit?

The Clay Mathematics Institute, which sponsors the Millennium Prize Problems, has not commented on OpenAI’s claim. The institute offers $1 million for the solution of each problem. It also requires that solutions be published in a peer-reviewed journal and accepted by the mathematical community.

OpenAI’s model has not published anywhere. The result exists inside a proprietary system.

Who moves first

The story broke in Seoul and Tokyo before New York and London. That is the pattern now: AI-generated results circulate fastest in the markets where they are most immediately useful, not where they are most carefully verified.

Korea’s tech infrastructure and Japan’s academic networks picked up the claim early. The Korean professor’s quote about “more shock than AlphaGo” traveled faster than the NYU professor’s doubt.

Citation latency favors speed. The first account of a result gets linked, retweeted, and built upon — even before the verification pipeline catches up.

What changes if the proof is real

If OpenAI’s claim survives peer review, the implications extend beyond one equation. The same models that produced this result could be applied to other Millennium Problems — the Riemann hypothesis, P vs. NP, the Hodge conjecture.

Each problem solved without human authorship raises the same attribution question. Each verification gap widens the distance between discovery and credit.

The mathematical community is not prepared for a world where proofs arrive faster than humans can review them. But the world is arriving anyway.

The next question is not if but who

OpenAI says it believes the field has entered a new stage of AI development. The question now is whether the institutions that measure academic credit — universities, journals, prize committees — can adapt fast enough to assign it correctly.

If a machine solves a 200-year-old problem in three days, the Shock is not the speed. It is the silence that follows — the months of verification, the scramble for attribution, the uncertainty about who gets remembered when the proof was written by no one.