OpenAI Forms Independent Math Advisory Group After Claiming AI Solved 100+ Open Problems

On September 21, 2026, OpenAI announced the formation of a nine-member Advisory Group on Mathematics and Artificial Intelligence, hosted at Princeton's Institute for Advanced Study. The panel's creation arrived alongside a sweeping new claim from the company: an internal AI model, which began its training run on August 28, 2026, has reportedly resolved more than 100 long-standing open problems across multiple fields of mathematics. This announcement represents a significant escalation in the ongoing intersection of artificial intelligence and formal reasoning, shifting the focus from single-problem demonstrations to bulk automated discovery. By establishing the advisory group at one of the world's most prestigious academic institutions, OpenAI is attempting to build a structural bridge to a mathematical community that has grown increasingly wary of corporate AI benchmarks.[1][2]

This development marks a rapid expansion from the company's earlier assertion on September 9 that its system had successfully solved the Navier-Stokes existence and smoothness problem—one of the seven Millennium Prize Problems established by the Clay Mathematics Institute—in just 88 hours of compute time. The pace of the subsequent 100 solutions reportedly surprised OpenAI's own researchers, prompting urgent internal debates on how to responsibly disclose the findings to the academic community without overwhelming the traditional peer-review system. Unlike previous AI milestones that focused on games or language translation, solving open mathematical problems requires rigorous formal verification. The sheer volume of the new claims has forced the company to acknowledge that dropping 100 unverified proofs onto the public internet could disrupt ongoing human research and complicate the attribution of mathematical discoveries.[1][3]

The newly formed advisory board includes prominent figures in pure and applied mathematics, deliberately selected to lend credibility to the review process. The founding members include Fields Medalists Timothy Gowers and Martin Hairer, alongside theoretical physicist Edward Witten and several other highly decorated academics. Operating entirely independently and without financial compensation from OpenAI, the group is tasked with reviewing emerging mathematical results produced by the company's internal models. Their primary mandate is to assess the mathematical significance of these AI-generated proofs, coordinate the timing and method of their public release, and advise on how these powerful new tools can be integrated into academic research without violating professional standards. By hosting the group at the Institute for Advanced Study, the initiative seeks to maintain a firewall between corporate product cycles and foundational scientific research.[1][2]

However, the board's mandate contains a strict and explicitly stated boundary regarding the company's internal research operations. OpenAI clearly noted in its announcement that the advisory group "will not be responsible for advising us on how to pace our internal progress on mathematics," ensuring that the academic advisors have absolutely no authority to slow the development or training of the underlying AI models. This carve-out means that while the mathematicians can control how and when the 100-plus solutions are communicated to the public, they cannot prevent the AI from continuing to generate new proofs at an accelerated rate. For competitors and academic observers, this structural limitation indicates that OpenAI intends to maintain its aggressive computational scaling, treating the advisory board strictly as a communications and verification filter rather than a regulatory brake on its research division.[1]

The creation of the advisory group follows mounting friction between AI developers and the traditional mathematical establishment. Earlier in September 2026, 25 Fields Medal laureates signed a highly publicized open letter warning that utilizing open mathematical problems merely as performance benchmarks for new AI systems could cause long-term damage to the discipline. The signatories argued that the fundamental goal of mathematics is conceptual understanding and the development of new theoretical frameworks, not simply the rapid generation of isolated solutions. They expressed deep concern that churning out automated proofs without human comprehension damages the collaborative nature of the field, shifting the focus toward brute-force computational output and away from the deep, intuitive insights that have historically driven mathematical progress.[1][2]

Notably, only one member of the new nine-person advisory group, Institute for Advanced Study mathematician Camillo De Lellis, was a signatory to that critical open letter. This composition suggests that the board represents a pragmatic middle ground—researchers willing to engage with corporate AI labs to manage the transition, rather than rejecting the technology outright. The panel's immediate and most pressing priority will be managing the rollout of the 100 claimed solutions, none of which have yet been published, itemized, or subjected to formal peer review by the broader academic community. Because OpenAI has not released the list of solved problems, mathematicians worldwide are currently operating in a state of uncertainty, unaware if the specific questions they have spent years researching have already been resolved by a proprietary algorithm operating behind closed doors.[1][3]

The broader mathematical community remains deeply divided on how to integrate these autonomous systems into formal research methodologies. While automated theorem provers like Lean have long been used to mechanically verify human-written proofs step-by-step, the generation of novel solutions at this unprecedented volume introduces an entirely different mechanism of discovery. It challenges traditional academic attribution by replacing human intuition with massive-scale pattern matching across high-dimensional vector spaces. If an AI system generates a valid proof that is tens of thousands of pages long and incomprehensible to a human reader, the community must decide whether that constitutes a true mathematical solution or merely a computational artifact. This tension highlights the growing gap between verifying that a statement is logically true and actually understanding why it is true.[2][4]

For now, the exact list of the 100 resolved problems remains a closely guarded secret. As OpenAI prepares to release the underlying proofs under the careful guidance of its new advisory board, researchers are waiting to see whether the AI's outputs represent genuine conceptual breakthroughs that can advance human understanding, or if the sheer volume of automated output will simply overwhelm the peer-review process. The coming months will test whether the Advisory Group on Mathematics and Artificial Intelligence can successfully bridge the cultural divide between Silicon Valley's demand for rapid benchmarking and academia's requirement for rigorous, comprehensible science. Until the proofs are published and independently verified, the true capability of OpenAI's new model remains an open question, suspended between a corporate press release and the rigorous demands of mathematical truth.[1][3]