Researchers Question OpenAI’s Handling of Unpublished Mathematics

A growing number of mathematicians are raising concerns over feeding unpublished proofs and working manuscripts into proprietary AI models like OpenAI's. At the heart of the debate is whether confidential mathematical ideas submitted for verification could be stored, incorporated into future training runs, or inadvertently exposed before formal publication. For researchers balancing the sudden utility of reasoning models against the protection of intellectual priority, the boundary between an assistant and a data liability remains precarious.