Overview
- OpenAI posted 722 AI‑generated manuscripts on GitHub on October 6 claiming progress on 372 open problems, including work tied to multiple Clay Millennium Prize questions.
- Less than 24 hours after the release the company withdrew three papers and issued multiple revisions after researchers found concrete errors such as a cascading sign mistake.
- Roughly half the documents lack machine‑checked formal proofs in systems like Lean, leaving most results dependent on months of independent verification and journal adjudication.
- Several mathematicians have accused the outputs of using unpublished human work without permission while OpenAI denies wrongdoing and the dispute remains unresolved.
- The release has produced sharp academic pushback and new national initiatives such as the Yau alliance in China that aim to build specialist AI math teams, which could reshape how math research is done and credited.