OpenAI forms independent mathematician panel to guide release of AI math results

OpenAI turns to elite mathematicians after bruising math rollouts

According to The Verge, OpenAI announced on Monday the creation of a nine-member Advisory Group on Mathematics and Artificial Intelligence (AGMAI), an independent body that will advise OpenAI and other AI companies on how they interact with mathematical research and the community around it — including how new results are presented and released. The outlet reports that the panel arrived abruptly and caught many mathematicians off guard, and that it follows a stretch in which the company's handling of spectacular mathematical results badly damaged its standing in the field.

A prominent roster, hosted in Princeton

The group is hosted by the Institute for Advanced Study in Princeton, New Jersey, and its members come from institutions including Stanford, Harvard, Oxford and Cambridge, counting multiple Fields Medals and MacArthur fellowships among them. Researchers The Verge spoke to described the lineup as the biggest names in the discipline. OpenAI says the panel will operate independently: it may offer advice the company never requested, comment publicly on the lab's impact on mathematics, and make its recommendations public. Members will not be paid by OpenAI, the group can change its own membership, and it appears it will be granted early access to OpenAI's research.

Murky origins, contested independence

The panel's sparse website says OpenAI initially approached some mathematicians about setting up an external advisory board, but that after talks with the company the researchers opted to launch an independent group instead and invite others to join. Which individuals were first approached, and how the remaining members were chosen, is unclear — a point Simon Machado of ETH Zurich raised with The Verge, describing the formation as opaque.

Group member Martin Hairer addressed the criticism in a blog post on Tuesday, pushing back on what he called widespread misinformation about how the group came together. He wrote that not every member had been contacted by OpenAI, that the resulting body is truly independent of OpenAI and other frontier labs, and that beyond confidentiality requirements members have signed nothing restricting what they may say publicly. He acknowledged the nine are hardly a representative sample of mathematics but insisted they are acting in good faith. The group is currently gathering input through conversations with colleagues, a public feedback form, and discussion on Proofs and Prompts, a blog that has become a hub for mathematicians working through AI's effect on their field.

An ivory tower?

Several researchers told The Verge the panel is a positive first step, but questioned whether nine elite academics can speak for the wider community. Francesco Fournier-Facio, an incoming professor at Heriot-Watt University, described himself as ambivalent: greater engagement with mathematicians can only help, but the group struck him as an ivory tower that does not reflect the researchers whose work and livelihoods are being upended as OpenAI pursues headline mathematical results. Machado made a similar point, saying the reality faced by mathematicians at that level differs sharply from the reality facing most working mathematicians today.

A large batch of results waiting in the wings

The panel's most immediate job is concrete. Hairer wrote that the group is currently advising OpenAI on how to coordinate the release of a large number of significant results the company says were produced by an internal model. Alongside the panel announcement, OpenAI said its unreleased model — the same one it credits with resolving the Navier–Stokes Millennium Prize problem — has now settled more than 100 long-standing open problems across most areas of mathematics.

Why it matters

The panel is a rare accountability experiment: a frontier AI lab inviting unpaid outside experts to review, and publicly challenge, how it announces scientific claims, at a moment when machine-generated results are colliding with the norms of attribution, priority and credit that hold mathematics together. Its effectiveness hinges on still-unanswered questions about its actual influence and representativeness, and on whether OpenAI heeds advice it did not ask for and may not like. With a reported backlog of major results queued up for release, that test could arrive quickly — and the arrangement may become either a template or a cautionary tale for how AI companies engage other research communities.