[REVISION HISTORY]

OpenAI math manuscript release and growing academic backlash

12 recorded revisions of this situation.

What changed

2026-10-22 12:42 UTC → 2026-10-29 22:12 UTC · added ~~removed~~

OpenAI has released 722 mathematical manuscripts on GitHub, generated by an unreleased internal frontier model. These findings, organized into 372 families of related results, span fields such as algebra, number theory, topology, and mathematical logic. OpenAI claims most results were produced from a single prompt given to a single AI agent, with each result requiring an average of approximately three hours of computing time. This release follows the company’s previous claim of solving the Navier-Stokes equations and includes reported progress on three additional Millennium Prize problems. The release has triggered significant debate regarding scientific rigor and transparency. While some researchers view the output as a major advancement, others remain skeptical due to a lack of independent verification. Notably, only 162 of the 722 papers include a main result formally verified by Lean software. Critics from MIT and the Institute for Advanced Study have highlighted that the claims cannot be replicated because the underlying model remains unreleased. Furthermore, the Institute for Advanced Study noted that OpenAI failed to disclose the specific prompts used, a move that contradicts recommendations made by its own independent Advisory Group on Mathematics and Artificial Intelligence (AGMAI). Backlash has intensified, with the Association for Human Mathematics (AHM) characterizing the publication as a “demonstration of power” rather than a “demonstration of scholarship.” Fields Medalist Terence Tao and others have called for a boycott of OpenAI. New technical concerns have emerged as OpenAI retracted three manuscripts due to sign errors. Additionally, researchers from the University of Cambridge and King's College London identified discrepancies between the AI's written explanations and the code used for computer-based verification via the Lean tool. In response to community friction, OpenAI withdrew its $1 million sponsorship of the Caltech Mathathon after 771 mathematicians signed an open letter criticizing the event.

Versions

Select a version to see what changed compared to the one before it.