Nine mathematicians have set up an independent advisory group on mathematics and artificial intelligence, hosted at the Institute for Advanced Study in Princeton, New Jersey. Its founding statement makes clear that the arrangement is not the one OpenAI originally proposed: the company approached some of the members about joining an external advisory board, and they chose instead to form a group of their own and invite the rest themselves.
Key takeaways
- The Advisory Group on Mathematics and Artificial Intelligence names nine initial members โ Timothy Gowers, Edward Witten, Martin Hairer, Ravi Vakil, Melanie Matchett Wood, Ulrike Tillmann, Camillo De Lellis, Franรงois Charles and Nikhil Srivastava โ none of whom are paid for the work.
- OpenAI says the same internal model behind its Navier-Stokes write-up has now resolved more than 100 further open problems across most areas of mathematics.
- The group holds no authority over how fast OpenAI pushes its mathematics research, and its first concrete task is advising on how that backlog of results should be released.
Why the mathematicians declined a company board
The distinction the group draws is structural rather than rhetorical. It says it operates independently of any AI company, that members take no payment, and that it will publish its recommendations on its own website rather than route them privately through a sponsor. It has also offered its advice to any AI company whose models are likely to move mathematics, not OpenAI alone.
What it does not claim is leverage. The group states plainly that it has no decision-making power at any AI company and that responsibility for each company's decisions stays with that company. OpenAI's own announcement draws the same boundary from the other side, saying the group is not there to advise it on how to pace internal progress in mathematics.
Within those limits the remit is broad: assessing the significance of emerging results, advising on how they are communicated and coordinated, weighing in on the academic and professional standards of mathematical research, and suggesting how AI tools might support mathematicians rather than displace them.
What triggered it
The group arrives in the middle of an open dispute. On September 11, twenty-five Fields Medal winners signed a declaration arguing that the race among AI labs to post solutions to famous problems is damaging the discipline. Compressed announcement cycles, they wrote, leave no room for write-ups or citation and raise severe attribution and plagiarism questions, and a correct answer alone is not what makes mathematical research valuable. The declaration proposed no rule and no enforcement mechanism, and it invited further signatories.
The overlap between that letter and the new group is thin. TechCrunch reported that of the nine initial members, only De Lellis had signed it. That gap is the interesting part. The people now formally advising OpenAI on disclosure are largely not the people who publicly objected to how disclosure has gone so far.
The immediate workload is concrete. The group says it has been asked to advise on coordinating the release of a large number of significant mathematical results that OpenAI reports its internal model produced โ the same model whose Navier-Stokes solution landed with an asterisk earlier this month. It has opened a form soliciting input from working mathematicians on how that release should be handled.
What to watch next
Because the group publishes its recommendations openly, the first real signal will be legible from the outside: a public recommendation on the release of the 100-plus results, followed by whatever OpenAI actually does. An unpaid body with no veto is easy to dismiss as decoration. A body that writes its advice down in public also creates a record of every time that advice is ignored.
The broader question it has put on the table is whether solving a problem and contributing to a field are the same act. That is not a question a benchmark score settles, and it is not one a large language model can settle on its own behalf.
FAQ
Does the Advisory Group on Mathematics and Artificial Intelligence work for OpenAI?
No. The group states that it operates independently of any AI company and that its members accept no payment for the work. It formed after OpenAI approached some members about an external advisory board, and it has said it is willing to advise other AI companies whose models significantly affect mathematics.
Can the group stop OpenAI from publishing AI-generated math results?
No. The group has said it holds no decision-making power at any AI company, and OpenAI has stated that the group is not responsible for advising it on the pace of its internal mathematics research. Its influence runs through published recommendations and the reputational weight of its members.
How many open problems does OpenAI say its model has resolved?
OpenAI says the internal model has resolved more than 100 open problems across most areas of mathematics, on top of the Navier-Stokes Millennium Prize solution published earlier. Those claims have not been refereed at the pace they were announced, which is a large part of why mathematicians objected.
