OpenAI’s Math Push Sparks Backlash as Rivalries and Credit Fights Shake the Field

OpenAI has claimed a major mathematical breakthrough by solving a long-standing Millennium Prize problem, but the announcement has also ignited a fierce dispute over credit, competition, and the company’s role in academic research. Mathematicians say the episode shows how AI giants are turning frontier math into a high-stakes race that could reshape the field for years.

The controversy centers on OpenAI’s rapid push into hard mathematics, its reported race against researchers at New York University and Anthropic, and accusations that the company pressured collaborators, blurred lines around data use, and treated academic work like a competitive trophy hunt.

For many in the field, the issue is bigger than one proof. It is about whether powerful AI companies are becoming so dominant that they can scoop researchers, absorb their work, and set the rules for who gets recognized when a difficult problem is finally solved.

What happened, and why does it matter?

OpenAI said this week that one of its unreleased models made progress on a Millennium Prize problem tied to the Navier-Stokes equations, a famously difficult set of questions about how fluids move. The company says the result came after a massive internal effort involving thousands of agents, millions of dollars in compute, and just 88 hours of work.

That would be an extraordinary achievement in any era. But in this case, the news landed amid simmering anger from mathematicians who believe OpenAI’s conduct highlights a growing imbalance between well-funded AI labs and the academic researchers who have spent years laying the groundwork for such advances.

The tension is also personal. Two mathematicians tied to the controversy, Tristan Buckmaster of New York University and Andreas Thom of the Technical University of Dresden, told The Verge that OpenAI’s behavior raised troubling questions about credit, collaboration, and whether the company’s tools were used in ways that were never fully disclosed.

How did a legendary math problem become a corporate race?

It started, according to OpenAI, with rumors. The company heard that outside researchers were making progress on one of the Clay Mathematics Institute’s Millennium Prize problems, a set of seven notoriously hard questions introduced in 2000 with $1 million attached to each solution.

Those problems are among the most prestigious challenges in mathematics. In the 25 years since they were announced, only one has been formally resolved: the Poincaré conjecture, a deep topological problem about three-dimensional spheres. That scarcity is part of what makes any breakthrough instantly valuable — scientifically, reputationally, and, increasingly, commercially.

OpenAI decided to test whether its own advanced model could make headway. It could, the company says. But the bigger story may be who it was apparently racing against.

At the time, Buckmaster and Levent Alpöge, a researcher affiliated with Anthropic, were also pursuing Navier-Stokes as part of separate research efforts. The race itself is not unusual in mathematics; priority disputes have always existed. What unsettled many observers was the scale and speed of OpenAI’s response once it learned rival researchers were close.

Buckmaster said the episode made clear to him that OpenAI was driven by one overriding aim: beating competitors to the result, not advancing the field for its own sake.

Why the Navier-Stokes problem matters

The Navier-Stokes equations describe the motion of fluids, from water flowing through pipes to air moving over wings. A full proof about their behavior would carry enormous theoretical weight, because the equations sit at the center of fluid dynamics, a field with relevance to engineering, physics, weather, and climate science.

For AI companies, solving such a problem is also a status symbol. It is a way to demonstrate that large models can do more than generate text or answer questions — they can contribute to frontier research once reserved for expert humans.

What are the Millennium Prize problems?

The Millennium Prize problems are seven open questions selected by the Clay Mathematics Institute to mark the start of the 21st century. Each carries a $1 million prize and each represents an area where the mathematical world has remained stuck for decades.

The seven problems span major branches of the discipline, and their difficulty comes from the fact that they are not classroom exercises with fixed methods. Researchers have to invent new tools, new perspectives, and often new subfields in the process of trying to solve them.

The Clay institute has said a solution must stand for two years after publication and receive broad acceptance from the global mathematics community before the prize is formally awarded. That means OpenAI’s reported breakthrough, even if valid, is currently in a kind of limbo.

The institute has removed Navier-Stokes from its list of unsolved problems, but it has not yet declared the question fully settled.

How did the dispute over Buckmaster and Alpöge escalate?

The conflict hardened around the issue of who was working with whom, and under what conditions. Buckmaster says OpenAI knew he was making progress and approached him with a proposal that, in his view, amounted to a pressure campaign.

According to his account, OpenAI researcher Sébastien Bubeck offered Buckmaster access to substantial company resources — including what Buckmaster described as effectively unlimited compute — if he would help finish the proof under OpenAI’s umbrella and take the paper’s lead authorship. But there was a condition that Buckmaster says made the offer impossible to accept: it would require dropping Alpöge from the effort because he worked for Anthropic, a direct competitor.

Buckmaster said he interpreted the offer as an attempt to buy silence and control the narrative around the result, and he rejected it outright.

OpenAI has disputed Buckmaster’s account. In a statement to The Verge, spokesperson Laurance Fauconnet said the company could categorically rule out the idea that Buckmaster’s Codex prompts over the past two months influenced the system in any way, including training. The company has also denied that anyone accessed his private user data.

Even so, the company has since acknowledged that it could not initially rule out the possibility that data derived from Buckmaster’s product use may have been used to improve the model — a caveat that has only deepened suspicion among some mathematicians.

Buckmaster later went public, saying his office became a sort of crisis center as colleagues helped review his work, manage communications, and speak with lawyers.

What did OpenAI say publicly?

OpenAI’s position has been that it did not improperly use Buckmaster’s specific data and that its breakthrough was the result of internal research and compute at scale. But the company’s comments also suggest a broader willingness to turn mathematical collaboration into a competitive exercise.

Bubeck, for his part, has denied Buckmaster’s description of the conversations in interviews and on social media. He has said OpenAI had similar arrangements with other mathematicians, although he did not identify them.

What Bubeck did acknowledge, however, is that Alpöge’s Anthropic affiliation was a serious complication. In his telling, OpenAI was unwilling to proceed with an internal project that included someone employed by a rival lab.

Why are mathematicians so alarmed?

Because they think the incentives are being distorted. Many researchers worry that AI labs do not value mathematical work in the same way academic mathematicians do. In academia, the point is often to expand the discipline: develop new ideas, establish new methods, and deepen understanding.

In a corporate setting, mathematicians say, the goal can look much narrower: be first, win the race, and claim the headline.

Tristan Buckmaster, who has also worked with Google DeepMind in the past, said he has seen a culture in tech research that prizes speed, prestige, and visibility above all else. In his view, the race to solve famous problems is less about discovery than about beating rivals to the finish line.

Buckmaster argued that in that environment, reputation becomes the only real currency, and the people doing the underlying science risk becoming secondary to the company that announces the result.

Andras Juhasz, a mathematics professor at the University of Oxford, described the discipline as part science and part art. Researchers are motivated by many things, he said, but often not by trophies. For many, the appeal lies in elegance, curiosity, and the joy of reaching a deeper level of understanding.

That difference in motivation matters because frontier mathematics is not a simple output machine. The route to a proof can be highly unpredictable, and the ideas that emerge along the way may matter as much as the final result.

Why credit is such a sensitive issue in math

In mathematics, credit is not just about status. It is about lineage. Researchers care who developed which tool, which idea unlocked which barrier, and how a new result connects to prior work. That lineage shapes the next generation of research.

When a tech company comes in and wraps a breakthrough in a product narrative, mathematicians worry that the field’s internal logic gets flattened into a simplified story about AI beating humans.

• Academic researchers want recognition for methods, not just headlines.
• AI companies want demonstrable wins that showcase model capability.
• That mismatch can create conflict over authorship, contribution, and disclosure.

What happened in Andreas Thom’s case?

Thom became part of the broader backlash after OpenAI announced another mathematical result that leaned heavily on his work and that of Gábor Kun. OpenAI later quietly updated its announcement to reflect their contributions, but without treating the correction like a major public event.

Thom told The Verge the experience was unpleasant, though he initially tried to move on. Buckmaster’s accusations prompted him to revisit a different concern: whether his own conversations with ChatGPT about research could have fed into the model improvements that later helped OpenAI make its breakthrough.

Thom said only OpenAI has the technical logs needed to answer that question. He also suggested the company may not have a clean answer itself, because different teams appear to have been using the system for different purposes without a clear understanding of where training data begins and ends.

Thom said the uncertainty itself was revealing, arguing that it suggests the company does not sufficiently care about the provenance of the intellectual labor feeding its systems.

He also said he resents what he sees as a failure to acknowledge the collective effort that made the breakthroughs possible in the first place. From his perspective, the companies are packaging a shared scientific inheritance as if it were simply their own product.

Could AI companies make scooping routine?

That is one of the biggest fears now spreading through the mathematics community. Traditional scooping — being beaten to a result — is familiar in science. But it has usually been limited by the fact that only a small number of people can work at the edge of a specific subfield, and doing so takes time.

AI labs change that equation. If a company can deploy thousands of agents, massive compute, and near-instant scaling whenever it hears a rumor that an important result may be within reach, then the old protections against being scooped start to disappear.

Researchers worry this could create an industrialized version of priority theft: one in which a powerful company can flood a problem with compute the moment someone else appears close to a breakthrough.

That possibility is already changing behavior. Some mathematicians told The Verge they are becoming more secretive about unfinished work. Others said they are reconsidering public lists of open problems because those lists could become roadmaps for corporate competition rather than useful scholarly tools.

What are the risks for young researchers?

The risks are especially serious for Ph.D. students and early-career faculty. These researchers often need time, stability, and visibility to establish themselves. If AI companies can repeatedly race ahead with far greater resources, the academic path becomes more precarious.

Shing-Tung Yau, the Harvard emeritus professor and Fields Medal winner who now teaches at Tsinghua University, warned that ambitious work is already risky for junior scholars. In his view, the arrival of well-funded corporate competitors could discourage people from taking on the most important questions at all.

That concern echoes broader battles across the creative industries, where writers, artists, musicians, and publishers have accused AI firms of building systems on large volumes of human work without permission or fair compensation.

How are mathematicians pushing back?

They are organizing, speaking out, and attempting to set norms before corporate incentives harden into default practice.

In June, a group of mathematicians released the Leiden Declaration, a set of principles for the responsible use of AI in mathematics. The declaration has since been endorsed by the International Mathematical Union and signed by thousands of people.

Its central message is caution: do not accept exaggerated claims from companies that overstate what their systems can do. That warning has taken on new urgency as AI firms begin presenting research-grade mathematical claims as evidence of superiority.

Resistance has also taken a practical turn. OpenAI withdrew sponsorship of an undergraduate mathematics hackathon at Caltech after pushback from people worried about corporate influence and the prospect of low-quality AI-generated submissions overwhelming the event.

Critics of the company said the dispute was not really about one hackathon or one proof, but about a broader pattern of tech firms trying to insert themselves into academic spaces without earning trust.

What happens next for OpenAI and the field?

OpenAI says it has already made substantial progress on another Millennium Prize problem, though the company has not disclosed which one. The speculation online ranges from the Hodge conjecture to other famous open questions, and rumors suggest Anthropic may be pursuing a major result as well.

That means the race is not slowing down. If anything, the current controversy may only push the labs to move faster and reveal less until they are ready to publish.

For mathematicians, the key question is whether the field can preserve its culture of openness, attribution, and careful verification while giant AI companies start treating open problems as strategic assets.

The answer may determine more than who gets a $1 million prize. It could shape how mathematical research is done, who gets to participate, and whether the next generation of scholars sees AI as a useful partner or an existential threat to the norms of the discipline.

Timeline of the controversy

Why this story reaches beyond one proof

Because the controversy is not only about whether OpenAI solved a famously hard equation. It is about whether the tools being built by the world’s richest AI companies are starting to reshape science in their own image.

Mathematicians are not merely asking who wins the race. They are asking who gets to define the race, who gets to benefit from the tools, and who is left out when the results are announced.

If the current pattern continues, some experts fear that the field will become more secretive, more defensive, and less collaborative. That would be a major shift for a discipline that has long relied on public problem lists, shared methods, and slow, cumulative progress.

OpenAI may well believe it is demonstrating the power of artificial intelligence to contribute to human knowledge. But the reaction from mathematicians suggests another interpretation: a warning that scale, speed, and competition can easily overwhelm the customs that make discovery possible in the first place.

For now, the company has a headline-worthy result and a community asking harder questions than ever about how it was achieved.