OpenAI Says Internal Model Resolved More Than 100 Open Math Problems as Independent Advisory Group Forms to Scrutinise Claims

OpenAI says a new internal artificial-intelligence model has resolved more than 100 long-standing open problems across multiple areas of mathematics, a claim that, if independently verified, would mark one of the most significant developments in the use of AI for fundamental research.

The company, in announcements reported on 22 September 2026, has also claimed that the system solved the Navier-Stokes problem, one of the seven Millennium Prize Problems designated by the Clay Mathematics Institute. OpenAI said training of the model began on 28 August.

However, the company has not yet publicly released the broader set of more than 100 claimed solutions for independent verification. Alongside the announcement, a new Advisory Group on Mathematics and Artificial Intelligence has been established to help evaluate and communicate such results.

An independent group of mathematicians

The advisory group, made up of prominent mathematicians, says it operates independently from AI companies, does not accept payment for its work and intends to publish its recommendations publicly. Its remit includes advising AI developers on how to interact with mathematical researchers and how to present and release potentially significant results responsibly.

The creation of such a group reflects the unusual challenges posed by AI-generated mathematical claims. Traditionally, new mathematical results are developed by researchers, written up in papers, circulated among experts and subjected to peer review before being accepted. The process can take months or years, particularly for complex proofs.

AI systems capable of producing large numbers of claimed proofs could overwhelm that process. Verification requires expert time, and the mathematical community must decide how to handle results produced by machines, how to attribute credit and how to ensure errors are detected.

Why Navier-Stokes matters

The Navier-Stokes equations describe the motion of fluids, from air flowing over an aircraft wing to water moving through pipes and blood circulating in the body. They are fundamental to physics and engineering, and are used extensively in simulations.

The Millennium Prize problem concerns whether smooth, physically reasonable solutions to these equations always exist in three dimensions, or whether they can develop singularities, points where the solution breaks down. Despite decades of effort by leading mathematicians, the question has remained open. The Clay Mathematics Institute offers a $1 million prize for a solution.

A claimed solution to a Millennium Prize problem is an extraordinary assertion. Only one of the original seven problems, the Poincaré conjecture, has been solved, by Grigori Perelman, whose work was verified by the mathematical community over several years. Clay's rules require that a proposed solution be published in a qualifying outlet and gain general acceptance in the mathematical community over time before a prize can be considered.

Caution is warranted

The history of AI in mathematics offers reasons for both excitement and caution. In 2025, AI systems from OpenAI and Google DeepMind achieved gold-medal-level performance at the International Mathematical Olympiad, demonstrating rapid progress in mathematical reasoning. AI tools have also assisted mathematicians in discovering new patterns and constructions.

At the same time, there have been episodes in which claims of AI-driven breakthroughs were overstated. In 2025, claims that an AI model had solved several open Erdős problems drew criticism after mathematicians pointed out that solutions already existed in the published literature, illustrating how easily apparent discoveries can turn out to be rediscoveries or errors.

Those experiences underline why independent verification is essential. Until the claimed solutions are published and scrutinised by experts, they should be treated as claims rather than established results.

What changes if the claims hold

If even a fraction of OpenAI's claims are verified, the implications would be profound. AI would move from being a tool that helps humans find information or check calculations to a genuine participant in fundamental research, capable of generating new knowledge.

For mathematics, that could accelerate progress on problems that have resisted human efforts for decades. It could also transform how mathematicians work, with AI systems proposing conjectures, generating candidate proofs and exploring vast spaces of possibilities, while humans focus on interpretation, verification and direction.

The implications would extend beyond mathematics. Many scientific fields, from physics and chemistry to economics and computer science, rely on mathematical reasoning. AI systems capable of original mathematical discovery could contribute to breakthroughs across disciplines.

Questions of credit and verification

The development also raises difficult questions. Who should receive credit for a proof generated by an AI system: the model, the company, the researchers who designed the system or the mathematicians who verify it? How should journals handle submissions of AI-generated proofs? And how can the community verify large numbers of claimed results efficiently?

Formal verification tools, which use software to check proofs rigorously, may play an important role. Proof assistants allow mathematical arguments to be expressed in a form that computers can check step by step. If AI-generated proofs can be translated into such formal languages, verification could become faster and more reliable.

The new advisory group's recommendations on responsible release and communication could help establish norms for these issues. Its independence from AI companies and commitment to transparency are intended to provide credibility at a time when commercial incentives may encourage bold claims.

The business context

The announcement also has commercial significance for OpenAI. The company is competing intensely with Google DeepMind, Anthropic and other developers to demonstrate leadership in advanced reasoning. Mathematical problem-solving has become a prominent benchmark for such capabilities because results can, in principle, be checked objectively. A verified breakthrough would strengthen OpenAI's position in that race and support its case to investors, enterprise customers and governments that its systems can contribute to high-value research.

That commercial dimension is precisely why independent evaluation matters. Companies have strong incentives to announce breakthroughs, while the scientific process depends on patient, sceptical verification. The advisory group's role in bridging those worlds could set an important precedent for how AI-generated discoveries are handled in other fields as well.

A moment of reckoning for AI and science

OpenAI's announcement arrives at a moment when AI companies are increasingly positioning their systems as tools for scientific discovery, not just productivity. Demonstrating original research capability would strengthen those claims and could influence how governments, universities and industry invest in AI for science.

For now, the mathematical world is watching and waiting. The next steps, publication of the claimed solutions, scrutiny by experts and the verdict of the advisory group, will determine whether this announcement is remembered as a turning point in the history of mathematics or as a cautionary tale about the limits of AI claims. Either way, the relationship between human mathematicians and machine intelligence has entered a new and consequential phase.