The claim is much larger than the published Navier-Stokes result
OpenAI said on 21 September that an internal model it began training on 28 August has now resolved more than 100 long-standing open problems across most areas of mathematics. The company says the same model produced its earlier Navier-Stokes result, which OpenAI published with a conventional manuscript and Lean formalisation and which the Clay Mathematics Institute subsequently described as an apparent settlement of the relevant Millennium Prize problem.
The new batch is at a different evidence stage. OpenAI has not published a catalogue of the more than 100 problems, the proposed proofs, formal certificates or an external validation record covering the set. The figure and the word resolved therefore remain OpenAI's account of internal results rather than an independently established mathematical outcome.
An independent group will advise on review and release
OpenAI has responded to the volume of claimed results by working with nine mathematicians who formed the Advisory Group on Mathematics and Artificial Intelligence. The group includes François Charles, Camillo De Lellis, Timothy Gowers, Martin Hairer, Nikhil Srivastava, Ulrike Tillmann, Ravi Vakil, Edward Witten and Melanie Matchett Wood. It is hosted through the Institute for Advanced Study and says it operates independently of AI companies.
The group says its immediate task is to advise OpenAI on how to coordinate the release of a large number of significant mathematical results that OpenAI reports were produced by its model. It plans to publish recommendations, accepts no payment from AI companies and can advise other AI developers. Its role is advisory: it has no decision-making authority over OpenAI or any other company.
Review becomes the bottleneck before scientific credit can be assigned
If OpenAI releases dozens of candidate proofs at once, mathematical significance cannot be inferred from the count alone. Each result needs its problem statement checked, prior literature searched, proof steps examined and, where formal artefacts exist, the formalisation connected back to the claimed theorem. Some problems may be famous and deep; others may turn out to have overlooked prior solutions, straightforward arguments or narrower interpretations than their labels suggest.
The advisory group's existence provides a channel for coordinating that process, but it does not itself validate the mathematics. OpenAI says the group can assess significance and advise on academic standards and dissemination, while the group says responsibility for company decisions remains with the companies. OpenAI also explicitly excludes the pace of its internal mathematics work from the group's remit.
Evidence boundary
The confirmed development is institutional: OpenAI announced the group, its members and remit, and the group independently describes its current work advising on a large body of results reported by OpenAI. TechCrunch separately reported the formation of the group and the company's 100-plus claim on 21 September.
The mathematical claim remains Reported. Until named problems, proofs or formal artefacts are released and reviewed by domain experts, there is no public basis for treating more than 100 additional open problems as solved. The useful next evidence will be the first tranche of published results, the advisory group's recommendations, and independent mathematical assessments of specific proofs.