OpenAI forms math advisory group as its AI resolves more than 100 open problems

OpenAI announced an Advisory Group on Mathematics and Artificial Intelligence, hosted at the Institute for Advanced Study in Princeton. The company says the same internal model that recently published a Navier-Stokes Millennium Prize solution has now resolved more than 100 additional open problems across most areas of mathematics.

Mathematicians are already pushing back on the sprint. Earlier this month, 25 Fields Medalists signed an open letter warning that AI labs are threatening intellectual work by racing to one-up each other on famous problems. A bridge to the community only matters if it can actually change behavior.

We’re split. Useful AI that cracks long-open math is an extraordinary scientific upside — if results are verified, attributed, and released without turning every seminar into a press race. Claiming 100-plus additional resolutions in the same breath as launching an “independent” advisory body is also a credibility stress test. Extraordinary claims need mathematicians in the loop before the champagne, not as a communications layer after.

Look at the charter carefully. Members aren’t paid; they can speak publicly, offer unsolicited advice, and control membership. They assess significance and help coordinate release. They do not get to pace OpenAI’s internal math progress — the blog says so explicitly, and IAS echoed that they have no decision-making power at any AI company.

So this is voice without a veto. That can still help: better release hygiene, clearer significance filters, less drive-by humiliation of working mathematicians. It cannot, by design, slow the lab if the lab doesn’t want to slow. Only one of the nine initial members — IAS’s Camillo De Lellis — also signed the Fields Medalists’ letter. That composition will be read as a signal, fairly or not.

Our desk wants both things at once: AI that expands what humans can prove, and norms that don’t treat the research commons like a scoreboard. Advisory theater that can’t touch the accelerator fails the second goal. I’m watching whether the group publicly challenges a release — and whether OpenAI ever lets advice change timing, not just packaging.

TechCrunch report by Aditya Mehta, Sep 21, 2026, on OpenAI’s announcement and IAS’s accompanying statement. Backdrop includes the recent Navier-Stokes publication controversy and the Fields Medalists’ open letter.