OpenAI's New Math Advisory Group Struggles to Prove Independence from Company

OpenAI appears caught in a familiar cycle: achieving genuine mathematical breakthroughs, then managing to mishandle their announcement. The company's latest attempt to reset relations with the mathematics community—by enlisting an independent advisory board—has itself become muddled, according to interviews with mathematicians and members of the new group.

The Advisory Group and Confusion Over Independence#

In late September, nine prominent mathematicians, including Fields Medal winner Terence Tao, announced the formation of the Advisory Group on Mathematics and Artificial Intelligence (AGMAI). The stated mission was straightforward: provide independent counsel to OpenAI and other AI labs on how to announce and evaluate mathematical research responsibly.

Yet OpenAI's own announcement of the group muddied the waters. While both OpenAI and AGMAI's statements emphasized the group's independence—claiming members could speak publicly and weren't paid by the company—much of the broader scientific community interpreted it as an OpenAI-created body. The confusion wasn't entirely unfounded: AGMAI's own website acknowledged that the group formed after OpenAI initially approached some eventual members about creating an advisory board.

Martin Hairer, an AGMAI member and mathematician at Imperial College London and ETH Zurich, tried to clarify the independence question in interviews and a separate blog post. He stressed that the group receives no funding or technical support from OpenAI and that members have signed only standard confidentiality agreements. "We don't work for OpenAI, are not paid by them, and it's totally independent," Hairer said in a video interview. Yet he also acknowledged frustration with how OpenAI had rolled out the announcement. "The way they phrased their announcement didn't really help," he noted, adding that while the company's wording was technically accurate, their framing was carefully designed to shape perception.

A Rushed Beginning Beset by Lessons Unlearned#

The entire initiative appears to have come together hastily. When asked about AGMAI's concrete plans, Hairer laughed and said, "We started two days ago. So it's not like we have a big master plan."

That urgency likely stems from OpenAI's eagerness to release an enormous backlog of results. Since solving the Navier-Stokes Millennium Prize problem—a high-profile achievement that subsequently descended into disputes over credit and methodology—OpenAI claims its internal model has resolved more than 100 additional major unsolved problems across mathematics.