Summary
On September 21, 2026, OpenAI announced that it was working with an independent Advisory Group on Mathematics and Artificial Intelligence, hosted at the Institute for Advanced Study in Princeton, to advise on how the company assesses and releases mathematical results produced by its models. In the same post, OpenAI said the internal model it began training on August 28, which it had earlier said resolved the Navier-Stokes Millennium Prize problem, "has now resolved more than 100 long-standing open problems across most areas of mathematics." The post cited the Fields medallists' declaration "A Severe Misalignment of AI in Mathematics" as the prompt, and stated that the group "will not be responsible for advising us on how to pace our internal progress on mathematics."
What Happened
OpenAI's post opened with the capability claim. It said the new internal model's "pace of its progress in mathematics has surprised the mathematicians within OpenAI," and that this "has led to internal discussions on the best way to inform the community of the rapid progress to prepare and adapt the field." The post did not list the problems, name the model, or say how the results had been checked.
It then referred to the open letter published by twenty-five Fields medallists on September 11, which it summarized as raising "concerns about the negative externalities of solving open problems as a benchmark for new AI systems." OpenAI wrote that the criticisms "highlight the need for thoughtful engagement of AI companies with the math community," and that it was "working with mathematicians who have established an independent mathematics advisory group" that "will serve as a bridge to the mathematical community and broader public, giving mathematicians a voice in how we move forward."
The post set out the group's remit and limits. It will advise OpenAI on assessing the significance of new results, on how "to coordinate their dissemination," on "academic and professional standards of mathematical research," and on how OpenAI's tools can support research and learning. It "will operate independently from OpenAI," may offer advice the company has not requested, may comment publicly on OpenAI's effect on mathematics and publish its advice; its members are not paid by OpenAI, and the group sets its own membership. Its remit excludes advising on the pace of OpenAI's internal mathematics work. OpenAI called the arrangement "a first step."
The nine initial members listed in the post are François Charles (ENS-PSL), Camillo De Lellis (IAS, GSSI), Timothy Gowers (Collège de France, Cambridge), Martin Hairer (EPFL, Imperial College London), Nikhil Srivastava (Berkeley, Simons Institute), Ulrike Tillmann (Oxford, INI), Ravi Vakil (Stanford), Edward Witten (IAS) and Melanie Matchett Wood (Harvard). Gowers, Hairer and Witten are Fields medallists; Hairer is the only member who was among the declaration's twenty-five initial signatories.
TechCrunch reported the announcement the same day, quoting the IAS's 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."
Why It Matters
The advisory group is OpenAI's institutional response to the medallists' declaration, which had asked the labs to address its concerns "urgently" without saying what that should mean. OpenAI's answer places outside mathematicians in charge of advising on how results are evaluated and released, which is the part of the dispute about credit, verification and the flooding of the field with announcements, and it excludes the question of pace, on which the declaration's argument about a benchmark race turns. The group's authority rests on publicity: it can publish its advice and criticize the company, but neither OpenAI nor the IAS gives it any power over decisions.
The capability claim in the same post is the larger unknown. The claim of more than 100 resolved open problems from a model that began training 24 days earlier rests, as of the announcement, on OpenAI's statement alone, with no list of problems, no papers and no external verification; the Navier-Stokes proof it builds on had not been verified by the community when the medallists published their declaration. How the group handles that backlog, and whether the results it reviews hold up, will show how much the arrangement changes the way machine-generated mathematics is released.
§ How to read the metadata
- Landmark
- Fundamentally alters the trajectory; 2–5 per year.
- Major
- Meaningfully shifts the landscape; 2–4 per month.
- Notable
- Worth documenting; significance can be upgraded later.
- Confidence
- High = primary sources corroborate. Medium = credible secondary only. Low = provisional. Disputed = credible sources disagree.
- Contestation
- Uncontested = no formal challenge. Contested = at least one challenge open. Superseded = replaced by a later entry. Unresolved = dispute still open.