OpenAI is setting up an advisory group for mathematicians to review and comment on the company's AI-generated mathematical research. The group, called the Advisory Group on Mathematics and Artificial Intelligence, will be hosted at the Institute for Advanced Study in Princeton, New Jersey. It arrives at a moment of tension between AI labs pushing to solve famous open problems and mathematicians who say the pace of those announcements is threatening their discipline.
The Navier-Stokes controversy that forced the issue
The announcement follows OpenAI's abrupt publication of a solution to the Navier-Stokes Millennium Prize problem, one of the seven unsolved problems with a million-dollar bounty. The release triggered immediate pushback from the mathematical community. Twenty-five Fields Medal winners signed an open letter arguing that AI labs are racing to one-up each other with solutions to famous problems, treating mathematical research as a competition rather than a collaborative endeavor built on careful verification.
OpenAI's response was to form a group that can review results and coordinate their release. But the company drew a sharp line around what the group can and cannot do. It will assess the significance of new results and advise on how they reach the public. It will not have any authority to slow down or redirect OpenAI's internal research. The blog post is explicit: "the group will not be responsible for advising us on how to pace our internal progress on mathematics."
What the group can actually do
The Institute for Advanced Study reinforced that limitation in its own press release. "Although we will give advice, we do not have decision making power at any AI company, and the responsibility for the decisions made by any company will rest with that company," the institute said. The group is advisory in the most literal sense. Members can offer unsolicited advice, go public with their views, and control their own membership. They cannot pause a release, demand more review time, or change how OpenAI approaches mathematical research.
Nine mathematicians have been named as initial members. Only one of them, IAS's Camillo De Lellis, also signed the Fields Medalists' letter. That composition suggests OpenAI chose members who are engaged with the intersection of AI and mathematics but not necessarily aligned with the most vocal critics of the company's approach.
The scale of the claimed results
As part of the advisory group announcement, OpenAI disclosed that the same internal model that worked on Navier-Stokes has resolved more than 100 additional open problems across most areas of mathematics. That is a remarkable claim. Open problems in mathematics are open for reasons. They resist solution because the existing tools and techniques are insufficient, or because the problem requires insight that no one has had yet. Claiming an AI has solved over a hundred of them in a single research push raises questions about the depth of those solutions, the standards of verification, and what "resolved" means in this context.
The mathematical community's skepticism is not purely about credit or pace. Mathematical proofs require human review. A solution that no one can verify is not a solution in any meaningful sense. The Fields Medalists' letter did not argue that AI should not do mathematics. It argued that the current approach, publishing results faster than the community can review them, undermines the process that makes mathematical knowledge reliable.
Advisory groups as a governance pattern
OpenAI's move fits a pattern emerging across AI labs. When faced with criticism about the impact of their research, companies form advisory groups that signal willingness to listen without ceding operational control. The groups have real experts, publish real opinions, and can speak publicly. They also cannot change timelines, redirect research priorities, or impose requirements on the company that hosts them.
For mathematicians, the question is whether this structure provides meaningful input or functions as a pressure valve. A group that can review results after they are published and comment on their significance is useful. A group that cannot influence when or how results are published is limited in its ability to address the core complaint, which is about pace and process, not just communication.
The tension between AI labs and mathematicians is not going away. As models become capable of more sophisticated mathematical reasoning, the question of how those results reach the public, who reviews them, and what standards they meet will keep surfacing. OpenAI's advisory group is one answer. Whether it satisfies the people it is meant to serve is a different question.