OpenAI has formed an independent advisory group of prominent mathematicians after researchers warned that rapidly improving AI systems could disrupt the way mathematical discovery is produced, evaluated and shared.
The company announced the group on Sept. 21, saying it will advise on the review and communication of emerging mathematical results, academic standards and the use of AI tools in research and education. The group is hosted at the Institute for Advanced Study and includes Timothy Gowers, Martin Hairer, Edward Witten, Ravi Vakil and other leading mathematicians.
The move follows an open letter published Sept. 11 by mathematicians who argued that using long-standing open problems as benchmarks for AI systems can work against the goals of mathematics itself. The letter says the field depends not only on obtaining answers but also on developing methods, understanding and new questions.
OpenAI says the debate has become more urgent because an internal model it began training Aug. 28 has resolved more than 100 long-standing open problems across mathematics. The company has also said the model produced a solution to the Navier–Stokes existence and smoothness problem, one of the seven Millennium Prize Problems. Those results remain claims by OpenAI and are separate from the advisory group’s role.
The Institute for Advanced Study describes the group’s purpose more broadly: to advise AI companies on their interactions with mathematical research and the mathematical community, including the responsible presentation and release of results.
Independent advice
OpenAI says the group will operate independently, may publish advice the company did not request and will not be paid by OpenAI. Members will also be able to criticize the company’s impact on mathematics publicly.
There is one significant limit. The group will not advise OpenAI on how quickly to pursue further internal progress in mathematics. That leaves the central question of capability pacing with the company even as outside mathematicians gain a formal channel for influencing how discoveries are reviewed and released.
The dispute is becoming an early test of a broader AI-safety question: what happens when systems become capable of producing valuable scientific results faster than professional communities can absorb, verify or integrate them?
AI Safety Watch 