OpenAI announced the formation of a Mathematics Advisory Panel composed of leading academic mathematicians, marking a formal step toward integrating AI tools into the frontier of pure mathematical research.
The advisory group was convened after OpenAI's latest reasoning models — believed to be iterations of its o-series — successfully resolved more than 100 mathematical problems that had remained open in the research literature. These problems spanned several branches of mathematics, including number theory, combinatorics, and algebraic geometry.
The mathematicians on the panel will advise OpenAI on how to responsibly deploy AI systems for mathematical discovery, evaluate the correctness and novelty of AI-generated proofs, and help identify high-value research directions where AI assistance could accelerate progress.
The announcement has divided the mathematics community. Proponents argue that AI can process vast combinatorial spaces and spot patterns that human intuition misses, potentially accelerating the pace of discovery in fields where progress has been measured in decades. Critics counter that resolving open problems is only valuable if the proofs are human-verifiable and conceptually illuminating, not merely outputs of a system no one can fully audit.
OpenAI acknowledged these concerns, stating that all AI-generated proofs undergo rigorous verification by human mathematicians before publication, and that the advisory panel's role includes establishing standards for what counts as a meaningful contribution to the field.
The development comes amid a broader trend of AI systems making inroads into scientific research. DeepMind's AlphaFold revolutionized protein structure prediction, and AI-assisted materials discovery has accelerated in recent years. But mathematics has long been considered a uniquely human domain — one where creativity and formal reasoning are inseparable — making OpenAI's claims particularly striking.
The panel is expected to publish its first set of recommendations on best practices for AI-assisted mathematical research by early 2027.
