OpenAI Claims AI Solved Million-Dollar Math Problem, Faces Accusations of Research Theft from Scholars
nashnova research
OpenAI says its internal model solved the $1 million Navier-Stokes Millennium Prize problem, but NYU mathematician Tristan Buckmaster alleges the company may have used his unpublished work without permission — turning the dispute from a math question into a battle over AI training-data boundaries and academic credit.
Did an AI really crack a 200-year-old math problem?
The Navier-Stokes existence and smoothness problem — whether the equations describing fluid motion always have "well-behaved" solutions — is one of the seven Millennium Prize Problems, carrying a $1 million reward.
On September 8, OpenAI announced that an internal AI model had solved it. This means → if true, it would be the first time AI independently cracked a top-tier pure-math conjecture — a milestone on par with AlphaGo.
But on the same day, NYU mathematician Tristan Buckmaster publicly alleged that OpenAI may have obtained his and his collaborator's unpublished research without authorization.
Why does the mathematician suspect his work was fed to the model?
Buckmaster and Anthropic employee Levent Alpöge had already made major progress on the same problem, using both Anthropic's Claude and OpenAI's Codex — and they entered their project drafts into Codex.
In plain terms = they fed their own solution approach directly into OpenAI's tool. If OpenAI's model ever "saw" that content, the claim of "AI solving it independently" becomes "AI copying someone's homework."
Buckmaster says he contacted a prominent OpenAI mathematician on September 3 to ask whether his Codex usage logs had been accessed. The response denied the model viewed user data, but his follow-up question about training data went unanswered.
What does OpenAI itself say?
In its announcement, OpenAI conceded: "While unlikely, we cannot rule out that de-identified data extracted from user product usage contributed to model improvements."
This means → OpenAI neither fully denied nor fully confirmed the possibility. The phrasing itself is a signal: they cannot prove the model's training data did not contain Buckmaster's work.
Altman separately admitted that OpenAI tried the problem because "rumors circulated online last week that an Anthropic model had solved a Millennium Prize problem." This reflects a competitive trigger, not pure scientific curiosity.
Why did OpenAI's two proposed "options" deepen the controversy?
Option A: Buckmaster and Alpöge publish their Euler-equation results one day before OpenAI's Navier-Stokes announcement; OpenAI cedes the prize to them, calling them "the humans closest to the problem."
Option B: Buckmaster publishes a Navier-Stokes paper alone, acknowledging that OpenAI's model solved it — but must remove Alpöge from the author list because Alpöge works at rival Anthropic.
Put simply = both options let Buckmaster keep his name, but require him to follow OpenAI's narrative timeline — while the Anthropic-affiliated collaborator is asked to "disappear." This reads less like academic negotiation, more like a business deal.
Can a closed-door briefing settle the dispute?
According to *Wired*, OpenAI held a closed-door media briefing on Tuesday. Executive Sébastien Bubeck denied accessing or using Buckmaster's Codex logs, stating: "Neither researchers nor AI agents saw any of their work before it was published last night."
Bubeck also acknowledged saying that things "would be simpler if Alpöge were not an Anthropic employee." This means → even without an explicit request to remove Alpöge, competitive dynamics were already shaping the authorship discussion.
Buckmaster framed the episode as a "Deep Blue vs. Kasparov moment," calling on the academic community to have a serious conversation about what comes next. All three parties have declined media comment requests. Whether this dispute receives independent verification will be the key test of OpenAI's claim.
市场有风险，内容仅供研究参考，不构成投资建议。