Advisory Group on Mathematics and Artificial Intelligence

On August 28, OpenAI says it began training a new internal model that has resolved the Navier–Stokes Millennium Prize problem and more than 100 long-standing open problems across most areas of mathematics. The company’s own mathematicians describe the pace as surprising.

That forces a community question larger than one lab release: if AI starts clearing research frontiers as a benchmark, what happens to careers, credit, verification, and the culture of proof? Mathematicians have already warned, in an open letter on “severe misalignment” of AI in mathematics, about negative externalities from treating open problems as scoreboards.

We’re treating the capability claim with care. OpenAI is asserting landmark results; independent verification and formal communication of proofs are exactly why an advisory bridge to the community matters. Extraordinary math claims need mathematicians, not press timing.

The Advisory Group on Mathematics and Artificial Intelligence, hosted at the Institute for Advanced Study, is framed as independent: unpaid by OpenAI, free to comment publicly, free to change membership, and explicitly not charged with pacing OpenAI’s internal math progress. Initial names include Timothy Gowers, Martin Hairer, Edward Witten, Melanie Matchett Wood, and others. That roster carries weight — if the independence holds in practice.

Our read lands in two layers. First, if the results hold, useful AI just touched one of science’s hardest domains. Novel mathematics can cascade into applications far outside pure theory; OpenAI is right that responsible deployment matters beyond the seminar room. Second, solving open problems as a private benchmark risks hollowing out the human pipeline that creates the next questions. A lab that clears a century of backlog without a dissemination plan can still harm the field’s incentives.

I’m watching whether the group gets real authority over review and communication — or becomes reputational cover while results drip out on OpenAI’s schedule. The contract OpenAI wrote for itself is revealing: advise on significance and standards, yes; advise on how fast OpenAI trains, no.

The trajectory if this becomes normal is AI-accelerated discovery with contested credit and a profession scrambling to redefine what counts as a contribution. That can be a renaissance or a gutting. Process will decide.

OpenAI announcement dated Sep 21, 2026, describing internal math progress since an Aug 28 training start and listing initial advisory-group members at IAS.