OpenAI released 722 AI-generated mathematics papers on October 6, 2026, before withdrawing three of them the very next day due to a sign error in an algebraic geometry proof. This error undermined the arguments of two related manuscripts that depended on it. The revised catalog now stands at 719 papers, with roughly 42 % formally verified using the Lean proof assistant. Critics point to the proprietary nature of the model, which prevents outside researchers from inspecting it or reproducing its process. The incident underscores that unrefuted claims with broken proofs remain, mathematically speaking, conjectures.
Source: Read the original article

