OpenAI's Math Advisory Group Faces Questions Over Independence and Trust

OpenAI's effort to build a new mathematics advisory body has run into familiar problems, according to reporting by The Verge. The Advisory Group on Mathematics and Artificial Intelligence, or AGMAI, was introduced on September 21 in a guest post on the blog of UCLA mathematician Terence Tao. It consists of nine mathematicians and is intended to advise OpenAI and other frontier AI labs on how emerging results are reviewed and communicated.

OpenAI stressed the group's independence, saying members could challenge the company publicly, publish their own advice, and offer guidance it had not requested. The company also said the group would not advise it on how quickly to pursue internal mathematics research. But the louder rollout created confusion. Many mathematicians, and most people outside the field, encountered AGMAI through OpenAI's announcements, and some initially assumed it was an OpenAI-appointed body. AGMAI's own website says the group formed after OpenAI approached some eventual members about setting up an advisory board, before those mathematicians opted to organize independently. It remains unclear which of the nine OpenAI initially approached.

Martin Hairer, an AGMAI member and a professor at Imperial College London and EPFL, told The Verge the group is independent. He said it receives no financial, technical or other support from OpenAI, and that no member has signed agreements restricting what they can say beyond standard confidentiality rules needed for early access to research. He described the preceding days as frustrating and intense, and said the group had only just started and had no large master plan. He said the group is open to working with other frontier labs and had already begun conversations, without naming them. Hairer was also critical of how OpenAI announced the relationship, saying the phrasing did not help, while acknowledging the company is careful with its wording.

That sensitivity has a context. The Verge reports that OpenAI's earlier mathematics announcements descended into disputes over credit, scooping and the treatment of researchers, leaving the community suspicious. Hairer acknowledged that AI companies may try to spin the advisory group's involvement to their advantage, and said his concern is the mathematics community rather than his own standing. He also said the potential disruption to researchers was one reason he joined a group of Fields Medalists addressing what they described as severely misaligned goals between AI companies and mathematicians.

OpenAI says its unreleased internal model has resolved more than 100 long-standing open problems across most areas of mathematics, and has hinted at progress on another Millennium Prize problem after its claimed solution to the Navier-Stokes problem became contentious. Álvaro Lozano-Robledo, a mathematics professor at the University of Connecticut, told The Verge that such statements do not match academic practice, since mathematicians either solve a problem or continue working on it. He said AI companies still need human mathematicians to judge how important a result is.

Researchers also complained about poorly written manuscripts and limited engagement with surrounding literature, which they said makes results hard to assess and can leave prior work uncredited. The Verge noted that OpenAI quietly changed documents after release without announcing the edits, and Hairer described alterations made in response to criticism as shoddy scholarship that breeds paranoia. Colva Roney-Dougal, a professor at the University of St Andrews, said the uncertainty is agonizing, and that the more results appear, the more likely her work or her students' work becomes irrelevant, leaving her unsure whether to rush papers, wait, or continue as normal. Some researchers told The Verge they fear a field they have spent decades on could be displaced, and suspect publicity motives ahead of an IPO.

The material does not establish when or how OpenAI's claimed results will be released, whether they have been independently verified, or how the advisory group will operate in practice. Those gaps matter as much as the claims themselves: the reporting describes a young panel trying to earn trust while the company it advises holds back a large set of unpublished findings.