OpenAI has announced an independent group of mathematicians to advise AI companies on how mathematical results should be reviewed, presented and released.
The Advisory Group on Mathematics and Artificial Intelligence is hosted at the Institute for Advanced Study and begins with nine members. It can publish recommendations, offer unsolicited advice, comment on OpenAI’s effect on mathematics and change its own membership.
Those are meaningful forms of independence. They do not give the group authority over OpenAI.
The group says it has no decision-making power at any AI company. OpenAI separately states that its remit does not include advising the company on the pace of its internal mathematics research.
That distinction matters because the announcement accompanies an extraordinary, unverified claim. OpenAI says a new internal model has resolved more than 100 longstanding open problems across mathematics, in addition to what the company describes as a resolution of the Navier-Stokes Millennium Prize problem.
The cited sources do not provide independent validation of those results.
What the group is supposed to do
OpenAI says the group will advise on the significance, review and communication of emerging mathematical results. It can also recommend academic and professional standards and consider how AI tools might support research and learning.
The group describes its purpose more broadly. It is willing to advise any AI company whose models could have a significant effect on mathematics, not only OpenAI. It says members will not accept company payment and that recommendations will be published on its website.
Its first task is narrower: advising OpenAI on the coordinated release of what the company reports is a large collection of significant mathematical results.
That could help establish a review sequence, give specialists time to inspect proofs and improve attribution and communication. It could also create a public record when the group disagrees with a company.
Independence does not mean control
The group came together after OpenAI approached some prospective members about creating an external advisory board. The mathematicians and OpenAI instead agreed to establish an independent group.
Its independence therefore concerns judgement, speech, membership and payment. The group can criticise OpenAI, publish advice that OpenAI did not request and work with other companies.
It does not have a veto over publication, deployment or research. Its own site states that responsibility for company decisions remains with each company. OpenAI also excludes the pace of its internal mathematical progress from the group’s remit.
This makes the group an independent adviser, not an independent regulator or approval body. Its influence will depend on whether companies provide enough evidence for meaningful review, respond to recommendations and explain when they choose not to follow them.
The 100-problem claim still needs validation
OpenAI says it began training the internal model on August 28, 2026, and that the model has since resolved more than 100 longstanding problems.
No list of all claimed results, common validation protocol or body of independent reviews appears in the announcement inspected for this article. A company report that a model has produced a solution is not equivalent to acceptance by specialists, journals, professional organisations or relevant prize authorities.
Mathematical validation can require more than checking a final proof for formal correctness. Reviewers may need to establish novelty, identify earlier work, assess hidden assumptions, isolate reusable ideas and turn a machine-produced argument into an explanation the field can understand.
The advisory group may help organise that process. Its existence does not itself validate any result.
Why mathematicians raised objections
The announcement follows a September 11 declaration titled “A Severe Misalignment of AI in Mathematics”, endorsed by numerous prominent mathematicians.
The declaration argues that treating famous open problems primarily as AI benchmarks can undermine the slower work through which mathematics develops concepts, methods, attribution and human understanding. It warns that rapidly producing answers without adequate exposition or integration could damage the research environment the systems depend on.
That is a broader objection than whether a particular proof is correct. It concerns who chooses problems, how results are credited, whether researchers can absorb a large release and whether solving a landmark problem advances understanding or merely records another completed task.
OpenAI cites the declaration as a reason for engaging the community. The advisory group is one response, but its limited authority leaves the main deployment, publication and research decisions with OpenAI.
What will demonstrate whether the group matters
The group’s public recommendations will be the clearest evidence of its independence in practice. Equally important will be OpenAI’s responses, including whether the company follows advice, delays releases when reviewers need time and discloses disagreements.
The other test is validation. The number of problems OpenAI claims to have solved is less informative than the quality, originality, review status and explanatory value of each result.
For now, OpenAI has created a channel for external mathematical judgement. It has not transferred control of its research programme, and it has not independently established the claimed breakthroughs.
