OpenAI's math advisory group launches into the confusion it was meant to fix

OpenAI has spent months pairing genuine breakthroughs in mathematics with badly handled announcements, and its attempt to mend relations with the mathematical community has begun in familiar fashion. According to The Verge, the company's latest move — supporting an independent advisory group of elite mathematicians — was rolled out in a way that left many researchers unsure what the group actually is.

A group born from OpenAI's approach

On September 21, nine prominent mathematicians announced the Advisory Group on Mathematics and Artificial Intelligence (AGMAI) in a guest post on the blog of Terence Tao, the UCLA professor and Fields Medalist who has been an outspoken critic of AI companies' conduct in mathematics. The group says it will operate independently and advise OpenAI and other frontier AI labs on the review and communication of emerging results.

OpenAI's own announcement leaned heavily on that independence: members would be free to challenge the company publicly, publish their advice, and offer guidance OpenAI had not requested. It also drew a boundary, noting the group will have no say in how quickly OpenAI advances internally on mathematics.

A confusing debut

According to The Verge, many mathematicians — and most people outside that community — first learned of AGMAI through OpenAI's much louder rollout, and confusion followed. Several mathematicians the publication spoke with initially assumed the group was an OpenAI-appointed body. That reading wasn't baseless: AGMAI's website says it formed after OpenAI approached some of its eventual members about establishing an advisory board, before they decided to organize independently and invite others to join. Which of the nine OpenAI originally approached remains unclear.

Martin Hairer, an AGMAI member and professor at Imperial College London and EPFL, told The Verge the group receives no financial, technical, or other support from OpenAI, and that members signed nothing restricting them beyond standard confidentiality requirements needed for early access to research. "We don't work for OpenAI, are not paid by them, and it's totally independent," he said. The group has already begun conversations with other frontier AI labs, though he declined to name them.

Hairer described the days after the announcement as "very frustrating" and "intense," and was candid about how improvised the effort remains: the group "started two days ago," he said, so there is no grand plan yet. Of OpenAI's announcement, he said the phrasing "didn't really help" establish the group's independence, while conceding the company's wording is technically careful. "They obviously are very good PR people," he said.

The backlog behind the rush

The urgency appears tied to what OpenAI is holding. The Verge reports that since its solution to the Navier-Stokes Millennium Prize problem — an announcement that itself descended into fights over credit, scooping, and the treatment of mathematicians — OpenAI says its unreleased internal model "has now resolved more than 100 long-standing open problems across most areas of mathematics." Coordinating the release of that pile is the first task handed to AGMAI, and the prospect is already stirring dread among researchers about what such a wave of results could do to their field.

Accusations of sloppy scholarship

Researchers who spoke to The Verge described OpenAI's recent mathematical output as poorly written, only lightly engaged with the surrounding literature, and difficult to evaluate — making the significance of results hard to gauge and leaving key work uncredited. They also pointed to documents being quietly revised after release to patch shortcomings, with no announcement or clear record of what changed. The Verge says it observed this around OpenAI's early-August announcement of "Ten advances in mathematics and theoretical computer science," and Hairer separately described instances of manuscripts being altered "sneakily" in response to criticism, calling it "really bad and sloppy scholarship."

Álvaro Lozano-Robledo, a mathematics professor at the University of Connecticut, put the cultural gap bluntly: "This is not how any academic behaves. I don't go around saying, like, 'Oh, I've proved all these things, but I don't know what to do with them.'" He drew a similar contrast with OpenAI's Laurance Fauconnet telling The Verge the company had "made substantial progress" on another Millennium Prize problem: "You either have solved it, or you're still trying."

Why it matters

The episode is about more than public relations. In mathematics, a solution and the understanding that comes with it have traditionally gone hand in hand, and AI-generated results threaten to pull them apart. If OpenAI really has more than 100 resolved problems waiting to be released, the work of reviewing, contextualizing, and crediting them will fall on a community that did not produce them — what Lozano-Robledo describes as a "burden" the field is being asked to carry. Quiet post-release edits make things worse, because mathematics depends on a stable, trustworthy record. Whether AGMAI can establish workable norms for publishing AI-generated results hinges on being seen as genuinely independent, and its muddled first week has made that task harder at precisely the moment the results are piling up.