OpenAI's math proof release falls short of advisory group's guidelines
OpenAI released hundreds of claimed solutions to difficult math problems after consulting an advisory group of mathematicians, but the results did not fully meet the group's guidelines. The Advisory Group on Mathematics and Artificial Intelligence had asked labs to stop testing advanced problems on proprietary models and to ensure human understanding of results. Only 10 of 719 manuscripts included the model's chain of thought, and many proofs were not formalized.
AGMAI is a nine-researcher body hosted by Princeton’s Institute for Advanced Studies. It published recommendations in late September. OpenAI said it consulted the group before releasing hundreds of claimed solutions to difficult mathematics problems.
OpenAI’s release still described evaluating proprietary systems on unresolved research questions, which conflicts with AGMAI’s first request. Only 10 of 719 manuscripts included the model’s reasoning chain. Forty-two percent of proofs had not been formalized. Separately, Cambridge and King’s College mathematicians found two mismatches between a natural-language proof and Lean code for a Navier-Stokes-derived problem.
The release may affect trust in AI-generated mathematical claims among mathematicians, students, and the public. Researchers could spend more time checking natural-language arguments against formal code, shifting labor toward verification. If labs follow advisory guidelines more closely, future results may become easier to audit and discuss. If not, unclear or unformalized proofs could spread confusion about what AI has actually solved. Funders and educators may also use such claims when judging AI’s mathematical abilities.