Layla Zulfa
OpenAI, the San Francisco-based artificial intelligence research laboratory, announced on Tuesday that its next-generation AI models have successfully solved one of the most elusive challenges in mathematics: the Navier-Stokes existence and smoothness problem. The announcement, which would normally mark a historic milestone in the annals of science, has instead been eclipsed by a burgeoning controversy involving allegations of intellectual property theft and professional intimidation. Shortly after OpenAI’s public declaration, Tristan Buckmaster, a prominent mathematician at New York University (NYU), came forward with a detailed statement alleging that OpenAI "played dirty" by leveraging his team’s private research to front-run the discovery.
The dispute centers on the Navier-Stokes equations, a set of partial differential equations that describe the motion of fluid substances such as liquids and gases. For over two centuries, these equations have served as the bedrock of fluid mechanics, used in everything from weather forecasting and ocean current modeling to the design of aircraft wings and blood flow analysis. Despite their utility, a fundamental mathematical question remains: do smooth, three-dimensional solutions always exist, or can the equations "blow up"? A "blow up" refers to a mathematical singularity where fluid velocity or pressure reaches infinity within a finite time—a scenario that is physically impossible but theoretically significant in determining the limits of the model.
In 2000, the Clay Mathematics Institute designated the Navier-Stokes existence and smoothness problem as one of seven Millennium Prize Problems, offering a $1 million reward for a definitive proof. OpenAI claims that its internal AI system, utilizing a multi-agent framework, has proven that the equations can indeed blow up under specific conditions, specifically within the context of "forced" Navier-Stokes equations.
The Mechanics of the Proof: 10,000 Agents and 88 Hours
According to technical documentation released by OpenAI, the solution was not the product of a single monolithic model but rather a coordinated effort by approximately 10,000 AI agents. These agents, which OpenAI describes as copies of a next-generation model significantly more advanced than the current GPT-6 Astra, worked in parallel to construct and verify the proof. The entire process reportedly took 88 hours of continuous computation.
To ensure the validity of the results, OpenAI utilized Lean, an interactive theorem prover and programming language. Lean allows mathematicians to encode mathematical statements and proofs in a way that can be verified step-by-step by a computer. By using Lean, OpenAI aimed to bypass the traditional, often years-long peer-review process required for such complex mathematical breakthroughs. The company claims the final output is a 100-page proof that provides a rigorous mathematical foundation for the "blow-up" phenomenon.
The use of "Astra," a model OpenAI suggests represents a leap in reasoning capabilities, marks a shift in how AI is applied to "hard" sciences. Unlike previous iterations of Large Language Models (LLMs) that relied on probabilistic guessing, the multi-agent system employed for Navier-Stokes appears to utilize advanced symbolic logic and recursive self-correction.
A Timeline of Conflict: Allegations of Plagiarism and Threats
The celebratory atmosphere surrounding OpenAI’s announcement was short-lived. Hours before the official reveal, Tristan Buckmaster of NYU released a four-page statement detailing a series of interactions with OpenAI leadership that he characterized as unethical.
According to Buckmaster, he and Levent Alpöge—a mathematician currently employed by OpenAI’s primary competitor, Anthropic—had been working independently for nearly a year on the same problem. Using a combination of custom AI tools and human intuition, the duo reportedly arrived at a solution for fluid blow-up by August 22. Buckmaster maintains that their project was a personal academic endeavor, unaffiliated with Alpöge’s employer, Anthropic.
The timeline of the alleged conflict is as follows:
- August 22: Buckmaster and Alpöge finalize their core findings regarding fluid blow-up.
- September 3: Amid industry rumors that Anthropic had achieved a mathematical breakthrough, Buckmaster contacted a mathematician at OpenAI to discuss his project, explicitly stating the work was personal and independent.
- September 6: Buckmaster participated in a conference call with Sébastien Bubeck, a high-ranking researcher at OpenAI. During this call, Bubeck reportedly informed Buckmaster that OpenAI’s internal model had already produced a 100-page proof for "forced" Navier-Stokes—the exact narrow approach Buckmaster’s team had been pursuing.
- The Ultimatum: Buckmaster alleges that Bubeck offered him a choice: either OpenAI would publish its results the day after Buckmaster’s team, or Buckmaster could publish alone, provided he removed Alpöge’s name from the paper. The latter condition was presumably intended to deny Anthropic any tangential credit for the discovery.
- The Alleged Threat: When Buckmaster refused to exclude his collaborator and insisted on a public disclosure of the timeline, he claims Bubeck asked, "Why would you ruin your career?" and stated, "If you don’t want me to be nice, then I don’t have to be nice."
Following this exchange, Buckmaster moved forward with the publication of his research papers, which credited both himself and Alpöge. OpenAI responded by releasing its own announcement, which Buckmaster views as an attempt to claim priority and overshadow his team’s work.
Official Responses and the Defense of Integrity
OpenAI’s leadership has moved quickly to refute Buckmaster’s claims. CEO Sam Altman issued a public statement defending the integrity of the research team. Altman asserted that the OpenAI team acted with "integrity and generosity" throughout the process and claimed that the company initially sought a joint release with Buckmaster’s team under the impression that both groups had solved the full Millennium Prize Problem.
Altman further clarified that OpenAI’s decision to publish independently came after realizing that Buckmaster’s team had primarily solved the Euler equations—a related but distinct set of equations for inviscid (zero viscosity) flow—rather than the full Navier-Stokes problem. "When we learned that the other team had only solved Euler, we realized our results were fundamentally different and more comprehensive," Altman stated.
Sébastien Bubeck also addressed the controversy, sharing excerpts of text messages that he claims prove OpenAI’s good intentions. Bubeck denied ever asking for Alpöge to be removed from authorship and maintained that OpenAI’s model had reached the solution independently.
However, OpenAI’s official statement regarding the use of user data remains ambiguous. While the company stated it had not viewed the pair’s work before publication, it admitted it could not "rule out that de-identified data derived from their usage of our products helped improve our models." This admission has sparked a broader debate about whether researchers who use AI tools to solve complex problems are inadvertently training the very models that will eventually compete with them for credit.
The Role of Formal Verification and Lean Software
The reliance on Lean software in this dispute highlights a significant shift in the mathematical community. Traditionally, a proof for a Millennium Prize Problem would be submitted to a journal, where it would undergo a grueling review process by a panel of experts. This process famously took years for Grigori Perelman’s proof of the Poincaré Conjecture.
By using Lean, both OpenAI and the Buckmaster-Alpöge team have moved the goalposts toward "formal verification." If a proof is written in Lean and the software’s kernel accepts it as correct, the mathematical community generally accepts it as logically sound, regardless of whether a human has read every page.
Terence Tao, a Fields Medalist and one of the world’s leading mathematicians, commented on the situation by praising the underlying math produced by Buckmaster and Alpöge as a "remarkable achievement." Tao’s endorsement adds significant weight to the academic side of the dispute, though he has yet to comment on the validity of OpenAI’s 100-page document, which has not yet been released for independent peer review.
Broader Implications for the Future of Scientific Discovery
The controversy at OpenAI is the second major AI-driven math story to emerge in a single week. Days prior, Anthropic announced that its Claude model had produced a computer-checked proof related to Fermat’s Last Theorem. This rapid succession of breakthroughs suggests that the "frontier" of mathematics is becoming a primary battleground for AI labs seeking to prove the "Reasoning" capabilities of their models.
The implications of this shift are twofold. On one hand, AI-assisted research could accelerate scientific discovery at an unprecedented rate, solving problems that have baffled humans for centuries. On the other hand, the centralization of these powerful tools within private corporations raises questions about the future of the "lone mathematician" and the traditional academic credit system.
If OpenAI’s proof is verified by the Clay Mathematics Institute, it would mark the first time an AI has solved a Millennium Prize Problem. However, the $1 million prize is typically awarded to individuals, not corporations or software agents. The institute has not yet commented on whether a "group of agents" or a corporate entity is eligible for the award.
Conclusion and Current Status
As of this week, the mathematical community remains in a state of cautious observation. Buckmaster and Alpöge’s most advanced result—the version closest to the official Millennium Prize Problem—is currently undergoing a final Lean check before publication. Simultaneously, the global research community is calling for OpenAI to release its full 100-page proof for independent scrutiny.
The fallout from this event is likely to influence the ethical guidelines surrounding AI in academia. The core of the grievance—the fear that AI companies are "scraping" the intellectual breakthroughs of their users to claim them as their own—poses a significant challenge to the collaborative spirit of scientific inquiry. Whether OpenAI’s Navier-Stokes solution is remembered as a triumph of technology or a cautionary tale of corporate overreach will depend on the transparency of the verification process in the months to come.
