OpenAI published 722 AI-generated mathematics papers on October 6, 2026, via its GitHub repository. The documents were produced by a proprietary model and addressed approximately 4,000 unresolved mathematical problems, including Millennium Prize conjectures.
On October 7, the company removed three papers from the collection. The withdrawal followed the discovery of a sign error in an algebraic geometry proof, which invalidated the logic in two dependent manuscripts.
The revised catalog now contains 719 papers. According to OpenAI, 300 of these results, or approximately 42%, have undergone formal verification using the Lean proof assistant software.
The project was developed with input from the Advisory Group on Mathematics and Artificial Intelligence. Critics have raised concerns regarding the proprietary nature of the model, which prevents independent verification of the process.
This incident follows a previous conflict between the company and the academic community, during which OpenAI withdrew a $1 million sponsorship for a Caltech AI mathematics competition following criticism from 771 mathematicians.
