Artificial intelligence is moving into one of the most demanding corners of research: advanced mathematics. OpenAI says an internal version of its next major model, Astra, produced results on 10 long-standing mathematics problems, including questions that had resisted progress for decades.
The reaction from mathematicians is not simple. The work points to a future where AI can help connect ideas across fields and accelerate discovery. It also raises uncomfortable questions about how human researchers are credited, how results are checked, and what happens to a discipline built around slow, expert-driven proof.
Astra’s mathematics results changed the conversation
James Maynard, a professor at the University of Oxford and winner of the Fields Medal, told The Verge that he has spent much of the past year “soul searching” over what AI could mean for mathematics. His concern reflects a broader shift: a field known for deliberate progress is now facing tools that may move much faster.
OpenAI said Astra tackled problems across a wide range of mathematical areas. Some were highly theoretical. Others had practical links to data transmission, error correction, quantum game theory, and post-quantum cybersecurity.
The list included work on how tightly spheres can be packed in more than three dimensions, which connects to efficient encoding and transmission of data. Another result involved error-correcting codes, which help recover information from noisy signals. Other work addressed when structural patterns emerge in complex connected networks, as well as the search for targets inside high-dimensional grids.
The core claim is not that AI is simply calculating faster. According to the source, the model can draw on patterns in training material and combine known results, methods, and tools in new ways. In mathematics, that can mean resurfacing ideas from academic literature or linking fields that are usually treated separately.
Why mathematicians are excited
Several mathematicians who spoke to The Verge viewed OpenAI’s announcement as meaningful, even when they could not personally assess every proof. That matters because the areas touched by Astra are highly specialized. A single researcher may understand one part of the collection but not be qualified to evaluate all of it.
OpenAI released more than 250 pages of papers explaining the solutions, along with 60 more pages describing “how the ideas came together.” The company also certified each result with Lean, software used to verify mathematical proofs.
That combination has helped give the work credibility. The problems were not presented as routine exercises. Maynard said they were the type of questions mathematicians and computer scientists had seriously studied and failed to settle.
Yang-Hui He, a fellow at the London Institute for Mathematical Sciences, put the significance plainly: “There’s a general feeling that [solving] one of these 10 problems would get you a job in academia.” He had recently returned from a four-week AI and mathematics research conference in South Korea, where he said many people felt the field had seen something like a “phase transition” over the past six months.
The credit dispute shows the risks
The most sensitive debate centered on non-sofic groups, infinite mathematical structures that roughly cannot be approximated by finite ones. Whether such groups exist had remained open for decades.
But the result also became a test case for how AI-backed research gives credit. Francesco Fournier-Facio, a mathematician at the University of Cambridge, told The Verge that people working in the area believed OpenAI’s first announcement played down the role of Andreas Thom and Gábor Kun, whose recent work helped support the result.
OpenAI initially described the collection as “results to problems that have been open and have seen no progress on the main result for at least a decade, and in most cases much longer.” Later, the wording changed to “results, each of which resolves or makes substantial progress on a long-standing open problem.” The source says the page did not include a correction note or an explanation for the change.
Kun, a researcher at the Alfréd Rényi Institute of Mathematics in Hungary, said OpenAI emailed him shortly before publishing. He found the original wording “rather comical,” because the fuller research paper “clearly said that it builds on my results from 2016 and 2019,” with the latter work coauthored with Thom. Kun’s assessment was blunt: “It’s rather sloppy.”
After publication, OpenAI contacted Kun again. He said an OpenAI mathematician wrote: “It was not intended to suggest that there had been no progress on this problem.” The email also said, “We certainly agree that the argument relies crucially on your work.”
OpenAI spokesperson Laurance Fauconnet confirmed that the language was updated. Fauconnet said OpenAI wanted the post to better reflect the prior research behind the results and to ensure those contributions were properly acknowledged.
Verification is only part of the challenge
Proof verification can help establish whether an argument holds. But this episode shows that correctness is not the only issue. Mathematics is also a human research culture, where attribution, context, and scholarly judgment matter.
That becomes harder when AI systems work across many specialized areas at once. OpenAI’s Astra results touched fields that few individual experts can evaluate in full. Even if the proofs are formally certified, researchers still need to understand where the ideas came from, which prior work mattered, and how much was genuinely new.
Kun said he wondered whether similar issues might exist in other results outside his expertise. That concern does not erase the possible achievement. It highlights the new burden that comes with AI mathematics: the field needs ways to check both the proof and the research record around the proof.
A field at the start of a major adjustment
The source describes a discipline already in upheaval. Mathematicians see the possibility of faster discovery, especially if AI can combine distant tools and surface overlooked connections. For difficult open problems, that could be powerful.
But the same progress creates anxiety. If AI systems can produce advances that once required years of expert work, the role of mathematicians may shift. Future researchers may spend more time guiding, interpreting, verifying, and contextualizing machine-generated arguments.
For now, the strongest conclusion is also the most careful one: OpenAI’s Astra has made a serious mark on mathematics, while exposing unresolved questions about how the field should handle AI-generated discovery. The takeover is not just about machines solving problems. It is about mathematics deciding how to absorb a new kind of collaborator.