OpenAI has claimed a historic victory in the world of pure mathematics, announcing that it has solved the Navier-Stokes existence and smoothness problem, one of the seven prestigious Millennium Prize Problems established by the Clay Mathematics Institute. The achievement, which the company says was accomplished in just 88 hours using roughly 10,000 AI agents and tens of millions of dollars in compute, should have been a cause for universal celebration. Instead, it has ignited a firestorm of controversy, exposing a deep and growing rift between the tech industry and the academic mathematics community—a rift defined by questions of credit, ethics, and the very soul of mathematical discovery.
The Mechanics of a Breakthrough: How OpenAI Solved the Navier-Stokes Problem
The Navier-Stokes equations describe the motion of fluid substances. While they are foundational to physics and engineering, proving that smooth, globally-defined solutions always exist (or that they can break down into a singularity) has remained one of the most intractable challenges in mathematics for over a century. OpenAI deployed an advanced, unreleased model to tackle the problem, deploying a swarm of agents that operated in parallel. The company states that it was motivated by rumors that independent researchers were nearing a solution, a competitive dynamic that would set the stage for the ensuing conflict. The solution itself has been submitted for publication and, in a move that has caused significant turbulence in the field, the Clay Mathematics Institute has removed the problem from its list of unsolved puzzles, though it has not yet formally declared it solved. The Institute’s rules require a two-year period of general acceptance within the global mathematics community before the $1 million prize is awarded.
The Cost of Victory: A Battle Over Credit and a Bribe Allegation
While the technical achievement is staggering, the narrative surrounding the discovery is dominated by human drama. Tristan Buckmaster, an NYU professor, and Levent Alpöge, a researcher at the AI company Anthropic, were reportedly close to completing their own proof of the Navier-Stokes problem. When OpenAI learned of their parallel work, the conflict began. Buckmaster claims that OpenAI researcher Sébastien Bubeck contacted him with an offer that he has publicly characterized as a “bribe.” According to Buckmaster, Bubeck offered him practically “unlimited compute” to finish his own proof and the opportunity to be the sole author of the paper announcing the breakthrough—a move that would have credited OpenAI’s tools but excluded Alpöge. “All I had to do was throw Levent under the bus,” Buckmaster told The Verge. He said he flatly rejected the offer.
OpenAI’s Defense and the Anthropic Problem
OpenAI has strongly denied the characterization of a bribe. Bubeck acknowledged in an interview that he offered OpenAI’s resources, but he framed it as an effort to collaborate and give Buckmaster the tools to complete his own work. Crucially, Bubeck admitted that Alpöge’s affiliation with Anthropic was a major obstacle. “From our perspective, how can we have an internal OpenAI project with an Anthropic employee?” he asked. This admission cuts to the heart of the issue: OpenAI saw the competition not just as a race for a mathematical trophy, but as a zero-sum battle between rival trillion-dollar companies. The mathematicians were caught in the crossfire.
When Did the Stakes Get So High? The History of the Millennium Prize Tensions
This is not an isolated incident. The tension between AI companies and mathematicians has been building for years. The Legend of the Millennium Prize Problems, set out by the Clay Mathematics Institute in 2000, includes seven of the most formidable challenges in the field. Only one—the Poincaré conjecture—had been solved in the quarter-century before OpenAI’s announcement. For an AI company, solving one of these problems is the ultimate trophy, a definitive signal of supremacy. For mathematicians, it represents the culmination of a lifetime of work, often built on the communal effort of generations.
Andreas Thom, a professor at the Technical University of Dresden, found himself at the center of a similar controversy last month. OpenAI announced a major mathematical result that heavily relied on work by Thom and fellow researcher Gábor Kun. The company quietly amended its announcement to acknowledge their contribution without a public announcement, a move that Thom described as “not a very pleasant experience.” These patterns—a rush to publish, a lack of transparency, and a perceived failure to properly recognize the intellectual lineage of the problems—are eroding the trust that is essential for the scientific enterprise to function.
What Are the Charges Against OpenAI? A Clear Breakdown of the Allegations
To understand the anger, it helps to isolate the specific, serious accusations being leveled by the mathematics community. While many are simmering, the core issues can be broken down into three distinct areas:
1. The Codex Data Question: Buckmaster had been using OpenAI’s Codex tool to tackle the Navier-Stokes problem. He has accused the company of failing to adequately explain whether material from his prompts was used to train or improve the models that ultimately produced the winning solution. OpenAI’s spokesperson denied this categorically, stating it was “impossible for Dr. Buckmaster’s Codex prompts over the last two months to have influenced the system in any way, including training.” However, Buckmaster remains deeply skeptical, citing the company’s behavior up to that point as justification for his distrust. The company later acknowledged that it could not rule out the possibility that data derived from his use of the product was used to improve the model in a more generalized way, a nuance that did little to quell concerns.
2. The Data Privacy Conversation: Andreas Thom’s concern goes a step further. He is asking the same question that many other mathematicians face: Could the conversations he and his colleagues had with ChatGPT about their research have been used to improve the models that ultimately cracked a problem they had spent years working on? Only OpenAI holds the data necessary to answer this question, and Thom suspects that the company does not even know. “To be honest, I suspect that they don’t even know,” he said. “Because it effectively means that they don’t really care, right?” This lack of clarity—this institutional indifference to the provenance of the data—is a profound violation of the trust that underpins academic collaboration.
3. The Conflict of Interest: A fundamental structural problem lies in OpenAI’s dual role. The company is both the provider of the primary research tools (like ChatGPT and Codex) and a direct competitor in the research field. This creates an unavoidable conflict of interest, especially when the company has privileged access to the unfinished, unpublished work of the mathematicians who rely on those tools. Fields Medalist Shing-Tung Yau has stated that this concern “deserves a substantive response. It should not simply be dismissed as ordinary competition.”
How the AI Race Is Causing a Chilling Effect on Mathematical Research
The most practical and damaging consequence of this controversy is the chilling effect it is having on the open culture of mathematics. The field has historically been a relatively open and collaborative environment, where researchers share preprints and present early-stage ideas at conferences, secure in the knowledge that the race to the finish is usually defined by deep, specialized expertise that is not easily replicated. AI companies change this equation entirely. With an industrial-scale ability to spin up thousands of agents and enormous compute power the moment a rumor of a breakthrough emerges, they represent a threat of a different magnitude.
Several mathematicians have told The Verge that colleagues are now reconsidering the practice of compiling lists of important unsolved problems. What was once intended as a valuable resource for the community is now seen as a potential target list for AI companies. Young researchers and Ph.D. students, who already face considerable risk in working on difficult problems, are now also contending with the possibility that a trillion-dollar company could sweep in and claim the prize, taking all the credit and funding with it. Yau warned that this prospect “could make them even more reluctant to pursue ambitious questions,” potentially stunting the growth of the entire field.
The Leiden Declaration and the Pushback Against AI in Mathematics
The academic community is not taking this lying down. In June, mathematicians published the Leiden Declaration, a set of principles for the responsible use of AI in mathematics. The declaration, which urges policymakers and the media not to buy into “the hype” created by companies that “overstate the capabilities of their products,” has been endorsed by the International Mathematical Union and has now been signed by nearly 3,900 people. The resistance is becoming more public and more organized.
The backlash also had a tangible impact on OpenAI’s own outreach. The company was forced to withdraw its sponsorship of an undergraduate mathematics hackathon at Caltech after fierce opposition. The opposition decried the intrusion of corporate interests and expressed worries that the event would create a deluge of low-quality “slop mathematics.” For a company eager to embed itself in the academic ecosystem, this was a significant public rebuke.
The core of the unease comes down to a single word: trust. Mathematicians do not have to accept the most explosive allegations to be worried about a company that both provides their research tools and competes with them. “I don’t really trust them,” Thom said bluntly. Buckmaster echoed the sentiment, noting that many of his colleagues are “actually scared” of the AI companies. He had planned to thank the colleagues who supported him publicly but chose not to after they expressed discomfort at having their names attached to the criticism. “The reality is that mathematicians are actually scared of them,” he said. Fear is a poor foundation for a productive and healthy research community, and Buckmaster is not convinced the companies care. “They don’t care anything about us as a community. It’s all about this petty drama between two trillion-dollar companies that are acting like children.”
The Clay Mathematics Institute’s decision to remove Navier-Stokes from its unsolved list places the problem in a peculiar limbo. The official $1 million prize and the formal declaration of a solved Millennium problem will not come for at least two years. Meanwhile, OpenAI has already moved on. In a statement, the company announced that it has made “substantial progress on another Millennium Prize problem,” though it has not officially disclosed which one. Unconfirmed speculation points to the Hodge conjecture, while rumors swirl that Anthropic is closing in on a problem of its own. As the two AI giants race to collect more of these intellectual trophies, they are learning that astonishing results alone are not enough to earn the trust of the community they are transforming. The chasm between their goals and the values of the mathematical world grows wider with every new announcement, leaving the field to wonder what the price of progress truly is.
