The question has moved from the realm of speculation to an increasingly urgent debate inside the mathematical community, as OpenAI prepares to release hundreds of solutions to previously unsolved problems.
People familiar with the company’s plans say OpenAI intends to publish the results on GitHub, potentially offering researchers access to a large body of machine-generated mathematical work. The release would represent another major step in the rapid expansion of AI-assisted mathematics, with frontier models producing tens of thousands of mathematical solutions this year.
The developments have generated both enthusiasm and unease among mathematicians.
OpenAI convened about 40 mathematicians in August to discuss how the field should respond if artificial intelligence begins to outperform humans at advanced mathematical research. According to people who attended the meeting, the company indicated that its increasingly powerful models had solved hundreds of long-standing mathematical problems.
The gathering was intended, in part, to seek advice about how such discoveries should be communicated to the scientific community.
Bryna Kra, a mathematician at Northwestern University who attended the meeting, recalls that the reaction was “a mixture of excitement and dread.”
For Kra and others, the central issue was not simply whether AI could solve difficult problems. It was whether the mathematical community would be given enough information to understand, verify and build upon those solutions.
She says attendees urged OpenAI not to simply announce major results through social media posts or corporate blogs, as the company had done with 10 mathematical problems earlier in August.
Instead, they argued that detailed scientific papers would be necessary so mathematicians could examine the methods, verify the claims and incorporate the discoveries into future research.
“Apparently, that input was ignored,” Kra says.
OpenAI disputes some aspects of the account. Spokesperson Lindsay McCallum says the company is “not aware” of the assurance described by attendees that the results would not be released all at once.
McCallum says OpenAI has been working to determine how best to publish its latest findings.
“On August 28, we began training a new internal model. In addition to resolving the Navier—Stokes Millennium Prize problem, this model has now resolved more than 100 long-standing open problems across most areas of mathematics,” McCallum says.
“We are working to responsibly release the next math results from our model, drawing on advice and public recommendations from the Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study to inform how we release these results. We have not set a release time.”
A scientific field caught in an AI race
For some leading mathematicians, however, the forthcoming release is about more than technological progress. It is also a test of whether AI companies are willing to respect the conventions that have governed mathematical research for generations.
Academics who spoke about the developments say they increasingly view mathematics as a competitive arena in which OpenAI and rival AI company Anthropic are demonstrating the capabilities of their models.
The concern is that the race to establish technological superiority could be moving faster than the scientific processes designed to verify discoveries, assign credit and preserve a reliable record of research.
One of the most contentious episodes occurred in September, when OpenAI deployed thousands of AI agents to work on a famous million-dollar Millennium Prize problem after reportedly hearing “rumors” that other researchers were close to a solution.
The episode drew criticism from Tristan Buckmaster, a mathematician and professor at New York University. Buckmaster said OpenAI had moved ahead on work he had been conducting in collaboration with Levent Alpöge, an Anthropic employee.
The pair had not yet published their work, although they had been using OpenAI's tools in their research.
A subsequent dispute concerned academic credit. Buckmaster claims OpenAI researcher Sébastien Bubeck appeared to suggest that Alpöge should not be included on a potential paper because doing so would complicate matters.
Bubeck denied asking for Alpöge to be excluded, pointing instead to a previous public statement in which he rejected that characterization.
Meeting notes reviewed by WIRED describe a tense conversation in which Bubeck expressed concern about Anthropic's activities and the possibility that the rival company could devote its computing resources to solving a Millennium Prize problem.
According to the notes, Bubeck said: “Levent must be talking to his leadership right now letting them know fucking OpenAI can get a Millennium Prize problem. What’s to stop Anthropic from giving all their compute to get a Millennium Prize problem?”
The dispute reportedly became more heated when Buckmaster threatened to speak publicly about his belief that OpenAI had taken advantage of his work.
According to the meeting notes, Bubeck responded: “If you don’t want me to be nice, then I don’t have to be nice.”
The episode has deepened concerns among some mathematicians about how the technology companies are entering a field built around careful attribution and verification.
Nestor Guillen, a visiting mathematics professor at NYU, says there is growing unease about the conduct of the AI industry.
“There’s a perception of mobster behavior” from the AI companies among mathematicians, Guillen says.
“We disagree with that characterization,” McCallum says.
But Guillen says the concern goes beyond artificial intelligence itself.
“I feel a lot of angst, and I see this more and more in my colleagues in mathematics, not over AI, but over the AI companies,” he says. “I think a lot of the angst is about the accumulation of power in one place.”
The problem with ‘math by tweet’
The disagreement has also exposed a broader tension between the speed of technological development and the slower traditions of academic science.
Mathematical discoveries are normally subjected to extensive checking and peer review. Researchers build upon earlier work, establish who contributed to a result and publish detailed proofs that allow other mathematicians to verify their conclusions.
AI companies, mathematicians argue, have often favored a different model: announcing striking results publicly and quickly, sometimes before the wider mathematical community has had an opportunity to scrutinize the work.
OpenAI and Anthropic have continued to publicize mathematical results through corporate blog posts rather than conventional scientific papers, according to mathematicians who spoke with WIRED.
Critics say this can make results difficult to verify and can obscure earlier contributions by human researchers.
The practice is not confined to AI laboratories. Alpöge himself announced through a social-media post that he had disproved an 87-year-old conjecture shortly after the World Cup final.
For Kra, however, the issue is not whether mathematical discoveries should be shared quickly. It is whether they are being shared in a manner that strengthens rather than weakens the research community.
“Math by tweet and math by press release to me is not the way to nurture the ecosystem that created the fertile ground that they have trained on,” Kra says.
The mathematical community has begun developing its own infrastructure to deal with the flood of machine-assisted research.
Among the new initiatives are Hexagon, a repository focused largely on AI-generated mathematical material, and Palomar, a registry for machine-verified mathematics. Such projects are intended to help researchers identify, organize and evaluate the rapidly growing volume of machine-generated work.
Kra says OpenAI has been encouraged to use these resources but has not substantially changed its approach.
“They haven’t changed their behavior,” she says.
She also argues that OpenAI's actions have not matched the principles laid out in the Leiden declaration, a call supported by more than 4,000 mathematicians urging AI companies to meet established standards within mathematics.
Is mathematics becoming obsolete?
Behind the immediate dispute lies a much larger question: if machines become better than humans at solving difficult mathematical problems, what role will mathematicians have?
People who have spoken with OpenAI employees say some within the company believe AI has effectively made mathematics obsolete.
McCallum rejects that interpretation.
“We don’t believe the future of mathematics is set,” she says. “We’re working with the math community to navigate the future collaboratively.”
The possibility that AI capabilities could overtake human mathematical researchers was presented as a hypothetical scenario at the August meeting. Yet some attendees interpreted the discussion as an indication that OpenAI believed such a transformation could arrive sooner rather than later.
One person familiar with the discussions compared the company's approach to the way “police might want to notify family before reporting a death in a car accident.”
Bubeck, meanwhile, has been described by people familiar with his thinking as believing that increasingly capable AI could rapidly eliminate many traditional mathematical careers.
Bubeck disputes that interpretation.
He says more powerful AI could instead allow mathematicians to pursue questions that would previously have been beyond their reach, connect mathematical research more effectively to practical problems and make the discipline accessible to a much wider audience.
“I see this as an opportunity to expand what mathematicians can do and the impact their work can have,” he says.
A field preparing to change, not disappear
For many mathematicians, the debate is therefore not about whether AI should be used. It is about how the technology should be integrated into a discipline that depends heavily on trust, transparency and accumulated knowledge.
Researchers recognize that they will need to adapt, particularly when it comes to preparing younger mathematicians for a profession in which AI may become an essential research tool.
Kra does not believe mathematics is approaching its end.
Instead, she sees the technology as an opportunity to redefine the boundaries of what mathematicians can attempt.
“It changes how we’re going to operate, but I think it’s a moment that we can think bigger,” she says.
She welcomes the possibility of having increasingly powerful AI systems available to help solve difficult mathematical problems. What she wants in return is greater transparency from the companies developing them—enough information for researchers to independently assess the results and confidently build on them.
The stakes, she says, are both unsettling and promising.
“It’s a scary time,” Kra says, “but it’s also really a deeply exciting time.”