OpenAI’s claim that its artificial-intelligence model solved the Navier–Stokes problem has ignited an intense scientific dispute over scholarly credit, data privacy, and the future of mathematical research. According to official announcements from the company on Sept. 8, 2026, an unreleased internal model solved the century-old fluid dynamics puzzle using about 10,000 AI agents that worked simultaneously for 88 hours.
The breakthrough immediately triggered friction within the academic community. Independent mathematicians argue that the system may have ingested their unpublished brainstorming sessions, raising deep attribution questions across creative and scholarly fields.
The Navier–Stokes Breakthrough and the Million-Dollar Prize
The Navier–Stokes equations, designated in 2000 by the Clay Mathematics Institute as one of seven Millennium Prize Problems, mathematically describe how fluids like liquids and gases move. The institute offers a US$1-million award for a verified solution, though it states it will only consider a proof valid after publication in a peer-reviewed journal and thorough community vetting.
OpenAI chief research officer Mark Chen called the achievement a significant milestone for artificial-intelligence research, stating that it proves even harder questions can eventually be answered. According to OpenAI, its GPT-6 Astra model took roughly 17 hours to verify the solution.
However, NYU professor Tristan Buckmaster and Levent Alpöge disputed the narrative. According to a statement released by Buckmaster, the pair spent the past year working on aspects of the fluid dynamics problem using tools from both OpenAI and Anthropic. Buckmaster alleged that OpenAI rushed to solve the problem after learning of their progress, utilizing a similar routing approach.
Attribution Controversies and AI Training Data
The core of the dispute centers on whether OpenAI models learned from proprietary brainstorming sessions uploaded by researchers. According to Luke McDonagh, who studies intellectual property law at the London School of Economics and Political Science, academic researchers may not fully grasp the consequences of uploading data and knowledge to personal AI accounts.
Buckmaster stated that he and Alpöge stored drafts in OpenAI’s Codex throughout their project. When rumours circulated that Anthropic had solved the problem, Buckmaster contacted an OpenAI connection, leading to a meeting with OpenAI researcher Sebastian Bubeck. During that call, Buckmaster asked whether the model had access to their Codex sessions. He stated that he was told the model did not look up user data, but did not receive a clear answer regarding training usage.
OpenAI strongly pushed back against the allegations. An OpenAI spokesperson told Nature and USA Today that, following an investigation, the company could confirm with full confidence that no user inputs past July 3 could have influenced the system. Bubeck added in a post on X that the lab began working on the problem due to viral rumors concerning Anthropic, and emphasized that the company used different approaches. OpenAI acknowledged broadly that de-identified user data from products like ChatGPT or Codex are used in general model training, but maintained that no specific user data was accessed to solve the Navier–Stokes problem.
The Clay Mathematics Institute designated seven Millennium Prize Problems in 2000, and until now, only one of the seven problems had been solved.
Broader Repercussions for Scholarly Credit
The tension extends beyond a single equation. Andreas Thom, a mathematician at the Dresden University of Technology, raised concerns that his brainstorming sessions with ChatGPT over the past year regarding non-sofic groups might have trained the chatbot. Thom noted that OpenAI posted a preprint in August reporting the first-ever example of a non-sofic group using a strategy similar to his own.
Thom noted that OpenAI’s paper correctly referenced earlier works by him and his collaborators, but because he had not opted out of training until late June, it remains impossible to know if his chatbot sessions assisted the model. He compared the interaction to talking through tricks of the trade with a human specialist who would normally credit those conversations in an acknowledgment section.
OpenAI did not provide a direct response regarding Thom’s sessions, but reiterated that users control whether their conversations help improve models and that opting out stops data use for model training.
What is the Navier–Stokes problem?
The Navier–Stokes equations describe how fluids and gases move, serving as a fundamental pillar of fluid dynamics. It is classified as one of seven Millennium Prize Problems by the Clay Mathematics Institute.
Why are mathematicians disputing OpenAI’s claim?
Researchers Tristan Buckmaster and Levent Alpöge alleged that OpenAI may have built upon their unpublished work and private brainstorming sessions without proper attribution. OpenAI called these claims categorically false, stating its researchers did not see the mathematicians’ work beforehand.
Has the solution won the million-dollar prize?
Not yet. The Clay Mathematics Institute stated it will only evaluate a solution after it appears in a peer-reviewed publication and undergoes rigorous community vetting.
Keep reading