OpenAI announced a breakthrough solution to the Navier-Stokes equations, one of mathematics' most famous unsolved problems worth a $1 million Millennium Prize, but the achievement is now overshadowed by allegations that the company may have plagiarized the underlying research and used proprietary data without consent. The controversy raises urgent questions about how artificial intelligence (AI) companies handle researcher data and whether corporate competition is eroding scientific ethics.

What Exactly Is the Navier-Stokes Problem?

The Navier-Stokes equations describe how fluids flow and behave. Mathematicians and engineers have been trying to prove whether these equations always have smooth, well-defined solutions for over a century. Solving this problem is so important that the Clay Mathematics Institute designated it as one of seven Millennium Prize Problems, each offering a $1 million reward to anyone who can solve it. OpenAI claims its AI models and research team have achieved this breakthrough, but the company has stated it does not intend to claim the prize money.

How Did the Plagiarism Allegations Emerge?

The story begins with two independent researchers: Tristan Buckmaster, a scientist at New York University, and Levent Alpöge, an employee at Anthropic (a competing AI safety company). According to Buckmaster, the pair spent roughly a year working on Euler's equations, a related mathematical problem that many consider a stepping stone to solving Navier-Stokes. They used large language models (LLMs), which are AI systems trained on vast amounts of text to generate human-like responses, as assistants to help with documentation and logic verification.

By August 15, 2026, Buckmaster and Alpöge had achieved their own breakthrough: proving "blowup results" for both the Boussinesq and Euler equations. They verified their proof on August 22 using Lean, a specialized programming language designed to check mathematical proofs for accuracy. However, Buckmaster noted that the AI-generated proof was "the most horrendous" he had seen, describing it as "AI slop," and he planned to rewrite it for clarity.

On September 3, Alpöge informed Buckmaster of rumors circulating that Anthropic had solved an important mathematical problem. Buckmaster then emailed an unnamed prominent mathematician at OpenAI to clarify that their work was a personal collaboration unrelated to either company. What happened next became the crux of the controversy.

What Are the Core Allegations Against OpenAI?

Buckmaster claims that OpenAI's approach to the Navier-Stokes problem was suspiciously similar to his team's strategy. He states that the specific idea of pursuing "forced blowup" in the equations was exactly what his team had "quietly" chosen, and that this direction "is not the direction one arrives at in a few days by giving a model the problem statement". This suggests that OpenAI may have had access to his team's research sessions.

During conversations with OpenAI's Sébastien Bubeck, Buckmaster asked directly whether OpenAI's Codex model (a code-generating AI tool) "had been trained on, or had access to, our sessions in Codex." OpenAI responded that Codex does not access user data. However, when Buckmaster asked whether the data was used for model training more generally, he apparently received no answer. This silence on a critical question has fueled suspicions about data practices.

OpenAI's own public statements on the matter appear contradictory. The company stated that "no specific user data was accessed in order to solve this problem," but also acknowledged it "cannot rule out that de-identified data derived from their usage of our products helped improve [its] models". This hedged language suggests the company may have used researcher data in ways that were not transparent or consensual.

What Happened When Buckmaster Refused OpenAI's Offers?

OpenAI allegedly presented Buckmaster with two options for how to proceed. The first was that Buckmaster and Alpöge publish their Euler proof first, after which OpenAI would publish its Navier-Stokes proof the following day, giving the researchers priority. The second option was that Buckmaster alone, without Alpöge, would co-author a paper acknowledging OpenAI's internal model had solved Navier-Stokes. Notably, Bubeck was adamant about excluding Alpöge, citing his employment at Anthropic as "annoying."

Buckmaster rejected both options and told OpenAI that if the company pursued the first option, he would go public with his concerns. This decision prompted what Buckmaster interpreted as a threat. According to Buckmaster's account, Bubeck asked him, "why [he] would ruin [his] career," and when pressed, responded, "If you don't want me to be nice, then I don't have to be nice". Bubeck then allegedly contacted Alpöge directly, questioning Buckmaster's sanity, though Alpöge refused to engage and redirected inquiries back to his colleague.

How Are the Two Sides Defending Their Positions?

OpenAI's perspective, as articulated by CEO Sam Altman, frames the company's actions as well-intentioned and cooperative. Altman supported Bubeck's decision to exclude Alpöge, saying "it was challenging to offer [the same publication options] to Levent". Bubeck himself justified the exclusion by arguing that "it would be inappropriate for an Anthropic employee to author OpenAI's work," a statement that some observers view as a double standard applied based on corporate rivalry rather than scientific merit.

Buckmaster and Alpöge, by contrast, maintain that their work was independent and that OpenAI's approach to the problem was too similar to be coincidental. They also emphasize that they funded the project themselves and used AI tools transparently as research assistants, not as a means to access proprietary methods.

Steps to Protect Research Data in the AI Era

• Audit AI Tool Usage: Researchers should carefully document which AI platforms they use and what data they input, then request transparency reports from those companies about how the data may be used for training or model improvement.
• Establish Clear Collaboration Agreements: When working with colleagues across institutions, create written agreements that specify intellectual property ownership, publication rights, and data handling practices to prevent disputes later.
• Request Data Deletion Confirmations: After completing research involving proprietary AI tools, explicitly request that companies confirm they have not retained or used your session data for model training, and obtain written confirmation.
• Use Privacy-Preserving Alternatives: Consider using open-source or locally-hosted AI tools for sensitive research rather than cloud-based commercial platforms that may have ambiguous data policies.
• Publish Preprints Early: Post research findings to preprint servers like arXiv as soon as they are ready, establishing a clear timestamp of discovery that can help prevent plagiarism claims.

What Does This Mean for the Future of AI-Assisted Research?

The Navier-Stokes controversy highlights a growing tension in academic and scientific research: as AI tools become more powerful and more researchers use them, the question of data ownership and usage becomes increasingly murky. Many researchers now routinely use LLMs like OpenAI's Codex and Anthropic's Claude as research assistants, treating them as tools for documentation, logic verification, and brainstorming. However, the terms of service for these tools often permit companies to use user data to improve their models, sometimes in de-identified form.

The fact that OpenAI could not or would not definitively answer whether Buckmaster's data was used for training suggests that current practices may not adequately protect researcher interests. This ambiguity creates a scenario where a researcher's own work could inadvertently contribute to a competitor's breakthrough, without consent or attribution.

"If you don't want me to be nice, then I don't have to be nice," Bubeck allegedly told Buckmaster, according to the researcher's account.

Sébastien Bubeck, OpenAI (as reported by Tristan Buckmaster)

OpenAI's proof of a finite-time blowup for the forced Navier-Stokes equations is technically a partial solution, addressing one of four possible categories established by the Millennium Prize criteria, rather than a complete solution to all scenarios. The proof will need to undergo peer review, a process that typically takes years, before any party can claim the prize or receive full scientific recognition.

The broader implications are significant. If AI companies can access researcher data through their platforms and use it to train models that then solve problems those researchers were working on, it creates perverse incentives and undermines the collaborative spirit of science. Researchers may become reluctant to use powerful AI tools for fear of having their work appropriated, which could slow innovation. Alternatively, researchers might demand stronger contractual protections, which could fragment the AI research ecosystem and reduce the benefits of shared tools.

As this dispute moves toward resolution through peer review and potentially legal channels, it will likely force AI companies to clarify their data policies and may prompt the scientific community to establish clearer norms around AI-assisted research. The stakes are high: not just a $1 million prize, but the integrity of the research process itself in an age of powerful artificial intelligence.