A Cold War Has Erupted Between the Mathematics World and AI
A disputed proof and whispered breakthroughs have triggered a silent standoff between frontier AI labs and the academic establishment.
By Aaron Rose · Tech Reader Magazine · September 17, 2026
The Announcement
The announcement landed before the mathematics world was ready.
OpenAI said one of its internal AI models had produced a possible solution to the Navier–Stokes problem, one of the hardest open questions in mathematics.
The company released a 166-page proof. It said the work showed that smooth starting conditions could lead to a singularity in a limited amount of time when an outside force was applied.
The full problem was not solved. The version without that outside force remained open. The proof had not been accepted by the Clay Mathematics Institute, which oversees the problem and its $1 million prize.
OpenAI said it did not intend to collect the money.
But the announcement had been made. The claim was now public. And almost at once, a disagreement over one proof grew into something much larger.
Two groups that had been moving toward the same future had collided.
On one side stood the mathematics establishment: universities, journals, independent researchers and professional groups built around human study and peer review.
On the other stood the AI laboratories: private organizations with powerful models, vast computing systems and the ability to attack difficult problems at a speed no university could easily match.
For a few days, their conflict became a hot war.
Then the open fighting cooled.
What remained was a cold war over the future of mathematics.
What remained was a cold war over the future of mathematics.
The Race Before the Announcement
Before OpenAI released its proof, New York University mathematician Tristan Buckmaster and Levent Alpöge, a mathematician working at Anthropic, had been studying the same version of the Navier–Stokes problem.
Their work used an approach first developed by mathematicians Diego Córdoba and Luis Martínez-Zoroa. Buckmaster and Alpöge had been working privately, using AI tools as part of their research.
Then word of their progress reached OpenAI.
What happened after that became the center of the dispute.
OpenAI committed major computing power to the problem. According to the company, thousands of AI agents worked in parallel. Its system produced the 166-page proof, and OpenAI announced the result before Buckmaster and Alpöge published their own work.
Buckmaster later raised serious questions.
When did OpenAI learn about the private research? Could information from his conversations with OpenAI’s Codex system have helped the company? How did OpenAI handle later talks about cooperation, authorship and credit?
OpenAI denied wrongdoing. It said its model produced the proof on its own. The company rejected the idea that it had taken private work from the researchers.
The two sides agreed on some basic facts. They did not agree on what those facts meant.
This was not simply a race to see who crossed the finish line first. It was a fight over where the race began, who had mapped the route and whether everyone had followed the same rules.
Those questions remain unsettled.
OpenAI denied wrongdoing. It said its model produced the proof on its own. The company rejected the idea that it had taken private work from the researchers.
A Claim Is Not a Prize
OpenAI could announce a proposed solution. It could not declare the proof accepted.
That power still belonged to the mathematics community.
Under the Clay Mathematics Institute’s rules, a proposed solution must appear in a qualifying publication. It must remain open for study for at least two years. It must also gain broad acceptance from experts before the institute can award the prize.
That process is slow by design.
A long and complex proof can hide a small error that destroys the entire result. Other mathematicians must inspect the work line by line, test its key claims and decide whether the argument holds.
OpenAI had the computing power to produce its proof.
But it still needed mathematicians to trust it.
That dependence became important as the conflict spread.
The Resistance Forms
The next battle did not take place over Navier–Stokes. It formed around an AI mathematics event at the California Institute of Technology.
Students had organized a Mathathon where participants would use AI systems to explore open problems. OpenAI and Anthropic were among its supporters.
To supporters, the event offered a look at the future. Students could use new tools to attack problems that had resisted older methods.
To critics, it looked very different.
Hundreds of mathematicians signed a public letter opposing the event. They questioned whether AI companies could gain ideas, skilled workers and public attention while students and researchers carried much of the risk.
The concerns went beyond one competition. Mathematicians were asking who would control the next stage of their field.
Would universities continue to guide mathematical research? Or would the largest discoveries move into private laboratories with closed systems and far greater resources?
OpenAI withdrew its support from the Mathathon, including computing credits it had promised.
Anthropic remained.
The withdrawal did not end the conflict. It showed how deep the conflict had become.
The mathematics establishment had drawn a line.
OpenAI had stepped back from it.
Then Scott Aaronson said aloud what others may have been discussing in private.
OpenAI withdrew its support from the Mathathon, including computing credits it had promised. Anthropic remained.
A Rumor
Aaronson is a theoretical computer scientist at the University of Texas at Austin. He has spent his career working near the border between mathematics, computer science and advanced computing.
On September 15, he published an essay titled “The Age of Wonders and Terrors.”
Aaronson described a mathematics community being overtaken by events. Longstanding problems were falling. More papers included statements explaining how AI had helped produce the results. Editors and reviewers were preparing for more machine-generated work than humans could examine by themselves.
Then Aaronson reported something more alarming.
He said he had heard rumors that AI companies already possessed solutions to other major open problems in theoretical computer science.
He did not claim they had solved P versus NP. He did not name the companies, identify the problems or reveal his sources.
He said the companies had been burned by the hostile reaction to the Navier–Stokes proof. According to what he had heard, they were holding back other results until they found a better way to release them.
The report remains unconfirmed.
But whether every detail of the rumor is true may not be the most important point.
The rumor itself suggests a situation that could be believed.
A respected computer scientist now considers it possible that private AI laboratories hold major discoveries they are not releasing. Mathematicians have reason to wonder what the laboratories know. The laboratories have reason to fear how the research community will react.
That is a cold war.
The Laboratories Go Quiet
The OpenAI dispute showed the risk of releasing a dramatic result.
A laboratory can spend millions of dollars and produce a proof in days. But once the proof appears, the company loses control of the response.
Researchers may challenge the mathematics. They may question the source of the ideas. They may reject the way credit was assigned. They may also see the announcement as a threat to their careers and institutions.
A result meant to display the power of AI can instead create suspicion of the company that produced it.
The other choice is secrecy.
A laboratory can keep a result private while its staff checks the proof, studies earlier research and plans how to release it. That may be careful and responsible.
It can also look like scientific knowledge being kept behind corporate walls.
The longer the silence lasts, the more questions grow.
What has been solved? Which model solved it? Whose earlier work made it possible? Have independent experts seen the proof? Is the result being delayed for scientific reasons, legal reasons or business reasons?
No one outside the laboratory can know.
The company avoids another public battle. But it deepens the cold war.
The Mathematicians Close Ranks
Academic researchers face their own problem.
AI systems can help them test ideas, search for useful paths and examine work that might take humans far longer to complete. Refusing those tools could leave researchers behind.
Using them may require trust that does not yet exist.
A mathematician working with a commercial AI system may enter unfinished ideas, private notes or early results. Depending on the service and its rules, that information passes through systems controlled by an outside company.
The Navier–Stokes dispute gave researchers a reason to think carefully before sharing.
Even if no private data was used improperly, the fear alone can change behavior.
Researchers may reveal less to AI systems. They may share less with colleagues who work inside private laboratories. They may keep early discoveries hidden until they are ready to publish.
Mathematics has always depended on competition, but it also depends on exchange. Researchers build upon earlier ideas. They discuss incomplete work. They test arguments with other people before publishing them.
A cold war rewards caution.
Caution produces silence.
A cold war rewards caution. And caution produces silence.
An Arms Race in Proofs
The two sides are not equal.
A university researcher may work with a small team and limited computing resources. An AI laboratory can run thousands of agents at once and spend millions of dollars on a single attempt.
But power in mathematics does not come only from speed.
A company can generate a proof without being able to grant that proof broad acceptance. It still needs skilled mathematicians to review the work, explain it and connect it to the rest of the field.
The mathematics establishment also cannot dismiss the laboratories.
If AI systems can solve problems that humans have studied for decades, universities cannot protect their role by pretending the systems are only clever assistants. The tools may be changing who—or what—performs the central act of discovery.
Each side holds something the other needs.
The laboratories have computing power, advanced models and the ability to search through possible solutions at enormous speed.
The mathematicians have professional trust, deep knowledge of their fields and the system that turns an interesting claim into accepted mathematics.
The laboratories can produce answers that may be difficult to trust.
The mathematicians can provide trust but may no longer control the production of answers.
That is not peace. It is mutual dependence under growing suspicion.
Each side holds something the other needs.
Two Worlds of Mathematics
If the cold war continues, mathematics may divide into two worlds.
The first would remain public and academic. Researchers would publish papers, attend conferences, train students and submit their work for peer review. The process would be slow, open and tied to institutions that have developed over centuries.
The second would exist inside private AI laboratories.
It would have larger computers, more powerful models and faster research cycles. It might hold results that no university team could easily reproduce. Its most important work could remain private until the company decided it was ready—or useful—to release.
One world would have public trust but limited power.
The other would have unmatched power but limited public trust.
Between them would sit a growing collection of questions no one had answered.
Can an AI system be an author? Who is responsible for a proof that no person fully understands? How should a company credit researchers whose published—or private—ideas helped guide a model? How can outside experts review a 166-page argument without becoming unpaid quality-control workers for a private laboratory?
And what happens when the AI systems produce proofs faster than human beings can check them?
The bottleneck would move.
The great challenge would no longer be finding the answer.
It would be deciding which answers to believe.
If the cold war continues, mathematics may divide into two worlds.
The World Outside the Standoff
The conflict may appear to be a fight over professional credit, research jobs and corporate status.
Its effects could reach much further.
Most people will never read an advanced proof. Yet mathematics supports work in computing, engineering, medicine, transportation, energy, communications and many other fields.
A discovery that looks purely theoretical today may help build an important technology years from now.
The public therefore has a stake in both speed and trust.
A valid result should not remain hidden because a laboratory fears criticism. A questionable result should not gain acceptance because a company needs a victory. Human researchers should receive proper credit, but professional status should not become a reason to block useful AI-generated work.
The public needs the laboratories to discover.
It needs the mathematicians to verify.
It needs both sides to communicate.
Instead, the researchers are becoming more guarded. The laboratories may be becoming more secretive. Each side is watching the other.
The people who could benefit from the discoveries are left waiting outside.
The public needs the laboratories to discover.It needs the mathematicians to verify.It needs both sides to communicate.
The Silence
The first battle ended without resolving who had crossed which line.
OpenAI released its proof. Buckmaster raised his questions. OpenAI issued its denial. Mathematicians organized against the Caltech event. OpenAI withdrew.
The open conflict cooled.
But the deeper struggle had only begun.
Now researchers must decide what they can safely share. AI laboratories must decide what they can safely release. Universities must defend their place in the discovery process, while companies race to prove that their machines can produce knowledge no human team has found.
And somewhere behind the closed doors of one or more AI laboratories, another major mathematical breakthrough may already be complete.
Or it may not exist at all.
The mathematics establishment cannot examine it. The public cannot benefit from it. The laboratories are not speaking.
For now, that silence is the clearest measure of the cold war.