OpenAI has announced the creation of an independent advisory board of leading mathematicians, aiming to avoid future reputational crises related to the presentation of mathematical results by artificial intelligence systems. The move comes after a series of incidents that have drawn criticism from the scientific community against the company, highlighting the boundaries between real scientific discovery and impressive — but unreliable — production of results from AI models .
See also: Apple challenges OpenAI's forensic analysis in trade secret case
The group, known as the Advisory Group on Mathematics and Artificial Intelligence (AGMAI) , is made up of nine distinguished mathematicians from institutions including Stanford , Harvard , Oxford and Cambridge . Its members include Fields Medal winners — the equivalent of the Nobel Prize in mathematics — as well as MacArthur “genius grant” recipients . The committee will be hosted by the Institute for Advanced Study (IAS) in Princeton , New Jersey , one of the world’s most respected centers for mathematical research, once the intellectual home of great figures such as Albert Einstein , John von Neumann and J. Robert Oppenheimer .
The announcement surprised many mathematicians, who acknowledge that it is a positive first step, but express serious reservations about the group’s real role and influence. Questions remain: how much power will the committee have? Will OpenAI to its recommendations? And can such a small group credibly represent the entire mathematical community?
Why OpenAI needs a mathematics advisory board
The issue is not new. Large language models ( LLMs ) like GPT-4o and OpenAI ’s o3 have repeatedly been shown to be capable of “solving” complex mathematical problems, while in practice producing results that look impressive but do not withstand rigorous mathematical scrutiny. This phenomenon — known as hallucination in the AI world — is particularly dangerous in mathematics, where a faulty proof can be confidently presented as valid.
The reputation crisis that forced OpenAI to act began when the company announced “spectacular” mathematical results from its models, which were then strongly disputed by the scientific community. Many researchers pointed out that OpenAI was presenting results without following the established standards of mathematical research — that is, without peer review, without transparency in methodology, and without adequate verification. This created a climate of distrust that the company is now trying to address.
AGMAI will operate independently, with the freedom to offer advice that OpenAI has not requested, to comment publicly on the impact of AI on mathematics, and to make its recommendations public. Members will not be paid by OpenAI , and the group will have access to the company’s research results before they are published. “ Its value depends on its members’ ability to exercise their judgment and challenge ours ,” OpenAI said .
See also: OpenAI Codex: Two sandbox boundary violations
Concerns about the independence of OpenAI's mathematics committee
Despite assurances of independence, many researchers remain skeptical. Simon Machado, a researcher at ETH Zurich in Switzerland, said the whole endeavor seems “shrouded in mystery” and questioned how the nine members were selected. The group’s website says only that it “came together after OpenAI approached some of its members about creating an external advisory body,” but that after discussions with the company, the mathematicians decided to establish an independent group and invite others to join.
The question of representation is central. The mathematical community is vast and diverse — it includes thousands of researchers from different countries, institutions, and disciplines. A group of nine people, even top-notch ones, can hardly represent this plurality. Moreover, the fact that the members come mainly from elite Western institutions raises questions about the geographical and thematic diversity of the group.
Another critical issue is the nature of the relationship between OpenAI and AGMAI. The company promises independence, but the very structure of the relationship — with OpenAI providing access to research data and setting the operating framework — creates inherent tensions. How can a committee be truly independent when it depends on the company for access to information? And what if its recommendations conflict with OpenAI?
OpenAI and mathematics: A persistent issue of credibility
Using AI in mathematics is one of the most ambitious — and controversial — areas of application for artificial intelligence. Companies like OpenAI, Google DeepMind , and Meta AI have invested significant resources in developing models that can help solve mathematical problems. Google DeepMind, for example, introduced AlphaProof and AlphaGeometry, systems that showed impressive performance on Olympiad math problems.
However, the difference between an AI system “solving” a problem and a real mathematical proof is enormous. Mathematics requires rigorous logic, complete transparency in the steps of the proof, and the ability to be verified by other researchers. An LLM that produces a “proof” may look convincing, but contain subtle logical errors that are not easily visible to non-experts.
OpenAI says that AGMAI will advise “ on the evaluation and communication of emerging results: helping OpenAI assess their significance, advising on coordinating their dissemination, and providing guidance on academic and professional standards for mathematical research .” The committee will also advise on how AI tools can support mathematical research and education — an issue with huge implications for universities and schools worldwide.
See also: Claude Opus 5 hacked OpenAI: Researchers entered its internal systems in 72 hours
Overall, the creation of AGMAI reflects a broader trend: AI companies are beginning to recognize that the credibility and trust of the scientific community are essential to their long-term success. If OpenAI wants its models to be widely used in research, it needs to demonstrate that their results are reliable and verifiable. The math committee is a step in that direction — but only if the company demonstrates in practice that it is willing to listen and act on its recommendations.
