OpenAI math advisory group AGMAI mathematics and AI

OpenAI Math Advisory Group: What It Can and Cannot Do

Nine leading mathematicians launched an independent AI advisory group on September 21, 2026, after OpenAI said its model solved 100+ open problems. Members, powers and limits.

m
· 6 min read · — views · — reads
OpenAI Math Advisory Group: What It Can and Cannot Do
Table of contents

OpenAI Math Advisory Group: What It Can and Cannot Do

Last updated: September 2026

At 1:15 PM Pacific time yesterday, September 21, 2026, TechCrunch reported that OpenAI had formed a math advisory group, and a new website, agmai.org, went live the same day with the full announcement published on Terence Tao's blog. The story behind the headline is bigger than a corporate board. Some of the most respected mathematicians alive have agreed to stand between AI labs and the mathematical community at the exact moment their collision became real.

The Advisory Group on Mathematics and Artificial Intelligence is an independent body of nine mathematicians, hosted at the Institute for Advanced Study in Princeton, that launched on September 21, 2026. It advises AI companies on handling mathematical research, after OpenAI said its internal model solved more than 100 additional open problems.

Header artwork of Terence Tao's blog where the announcement was published

Source: Terence Tao's blog

Who the nine founding members are

The group did not emerge from a corporate press release. It was announced by a member, on a mathematician's blog, which is itself a statement about how this community communicates.

  • François Charles — ENS-PSL
  • Camilo De Lellis — IAS and GSSI
  • Timothy Gowers — Collège de France and Cambridge
  • Martin Hairer — EPFL and Imperial College London
  • Nikhil Srivastava — Berkeley, Simons Institute
  • Ulrike Tillmann — Oxford and INI
  • Ravi Vakil — Stanford
  • Edward Witten — IAS
  • Melanie Matchett Wood — Harvard

Read that list again slowly if you follow mathematics at all. It spans several of the world's top departments on two continents, and it includes mathematicians known for work that shaped modern analysis, combinatorics, and mathematical physics. For an international student preparing for a mathematics degree, these are names that appear in textbooks and lecture notes, now taking positions on how AI should treat their field.

Fuld Hall at the Institute for Advanced Study, Princeton, host of the new group

Source: Wikimedia Commons

Why this happened now

Two events in September 2026 made this group necessary rather than theoretical. The first was the sudden publication of a claimed solution to the Navier–Stokes problem, one of the Clay Millennium Prize problems, which we covered in depth in our Navier–Stokes solution report. The second was OpenAI's statement, reported by TechCrunch, that its internal model had solved "more than 100 additional open problems" across most branches of mathematics.

The mathematics community did not take this quietly. Earlier this month, 25 Fields Medalists signed an open letter criticizing AI labs for racing to solve famous problems in ways they saw as threatening mathematicians' intellectual work. Only one member of the new advisory group, Camilo De Lellis, was among the signatories. That detail matters: this is not the protest faction setting up shop, but a bridge that includes people sympathetic to both sides.

The origin story, as told on agmai.org, is unusually organic. OpenAI approached some of the members about creating an external advisory board. Together, they decided that an independent group, free to advise any AI company and not just one, would be more credible and more useful. They then invited others to join. OpenAI, in a statement quoted by TechCrunch, said: "This group will serve as a bridge to the mathematical community and broader public, giving mathematicians a voice in how we move forward."

What the group can and cannot do

Here is where honesty matters, because the announcement could easily be misread as mathematicians gaining power over AI labs. They have not.

The group can The group cannot
Advise any AI company whose models affect mathematics Slow or redirect OpenAI's ongoing mathematics research
Publish its recommendations publicly on agmai.org Make or veto decisions inside any AI company
Choose its own members and set its own agenda Receive payment for its advisory work

The limits are stated plainly on agmai.org: the group has no decision-making authority in any AI company, and responsibility for decisions rests with the company itself. TechCrunch reported the same boundary from OpenAI's side, noting that the group will not be granted the power to slow or redirect the company's ongoing mathematical research, and quoting OpenAI directly: the group "will not be responsible for advising us on how to pace our internal progress on mathematics."

So the influence is real but soft. The group's leverage comes from reputation, public recommendations, and the fact that AI labs currently need mathematicians to check, contextualize, and legitimize machine-generated results. Its immediate task is coordinating the release of a large number of significant mathematical results that OpenAI says its internal model produced, which is precisely the kind of high-stakes publication where presentation decides whether the community learns or merely reacts. For students and researchers watching from Lagos, Chennai, or Jakarta, this is the working answer to what an advisory group without authority can actually accomplish.

While the institutions sort out governance, the tools are already changing daily research. Formal verification is growing fast, as we explained in our Leanstral 1.5 formal proofs walkthrough, and Truescho's free AI study tools can help you build the proof-writing habits this new era rewards, alongside scholarships and university rankings for planning your next degree.

What this means for you

If you are a mathematics student anywhere in the world, three practical consequences follow from yesterday's announcement. First, expect a wave of machine-generated results to enter the literature over the coming months, released under the coordination process this group now runs. Learning to read such results critically, checking their formal verification status rather than their headlines, becomes a core academic skill.

Second, the boundary between using AI and doing mathematics is being redrawn in public. The Fields Medalists' letter and this group's creation are the two sides of that debate, and graduate admissions committees, supervisors, and journals will all be watching how OpenAI's next releases are presented. Following the group's public recommendations on agmai.org is probably the cheapest research-literacy upgrade available right now.

Third, the story is a reminder that AI capability and AI governance now move together, a pattern we saw as well in our coverage of the GPT-5.6 launch. For anyone planning a research career, understanding both the models and the norms forming around them is the equivalent of learning the rules of a sport you intend to play professionally.

Frequently asked questions

What is the Advisory Group on Mathematics and AI?

AGMAI is an independent group of nine mathematicians based at the Institute for Advanced Study in Princeton. Launched on September 21, 2026, it advises AI companies on the responsible presentation of mathematical results and on how their models affect the mathematics community. Its recommendations are published openly on agmai.org.

Why did OpenAI create a math advisory group?

OpenAI approached some mathematicians about an external advisory board, and together they chose a fully independent group instead. The immediate task is coordinating the release of a large batch of significant results that OpenAI says its internal model produced, after the model solved more than 100 additional open problems.

Who are the members of the AGMAI group?

The nine founding members are François Charles, Camilo De Lellis, Timothy Gowers, Martin Hairer, Nikhil Srivastava, Ulrike Tillmann, Ravi Vakil, Edward Witten, and Melanie Matchett Wood. They come from institutions including the IAS, Oxford, Cambridge, Stanford, Berkeley, Harvard, ENS-PSL, EPFL, and GSSI.

Did OpenAI's AI really solve 100 open math problems?

According to OpenAI, as reported by TechCrunch on September 21, 2026, its internal model solved "more than 100 additional open problems" across most branches of mathematics. The claim has not been independently verified, which is exactly the kind of announcement the new advisory group wants to handle responsibly.

Is the advisory group independent from OpenAI?

Partially. The group is unpaid, publishes its recommendations publicly, and can advise any AI company. But OpenAI stated that the group "will not be responsible for advising us on how to pace our internal progress on mathematics," so it cannot slow or redirect the company's research.

What was the Fields Medalists' open letter about?

Earlier in September 2026, 25 Fields Medalists signed an open letter criticizing AI labs for racing to solve famous problems and threatening their intellectual work. Only one advisory group member, Camilo De Lellis, was among the signatories. The letter is key context for this week's announcement.

What to watch next

Three signals will tell you whether this experiment matters. Watch agmai.org for the group's first public recommendations, because their tone will define whether "advisory" means influential or ceremonial. Watch how OpenAI releases the batch of results the group is now coordinating, and whether each comes with the formal verification that tools like the one in our Leanstral 1.5 report make possible. And watch whether other AI companies accept the group's standing offer to advise anyone whose models touch mathematics, which would turn a bridge into genuine infrastructure. For now, the most useful thing any student can do is stay informed and keep building skills: Truescho's free platform brings together AI study tools, scholarships, GPA planning, and 1500+ university rankings to help you do exactly that.

Sources