Overview
- OpenAI published 722 AI‑generated preprints to GitHub on October 6, saying they made progress on 372 open problems across algebra, geometry and logic including items that touch three Millennium Prize challenges.
- Less than 24 hours after the release a single sign error led OpenAI to withdraw three papers and post at least 14 revisions, showing immediate instability in the set of claims.
- Leading mathematicians and groups have organized formal pushback with Terence Tao and the Association for Human Mathematics urging a boycott and more than 8,000 researchers signing an open letter that the release was rushed and unvalidated.
- Roughly half the manuscripts were published without Lean machine‑checked formal proofs and OpenAI has not released its internal model, full prompts, or training provenance, which makes independent replication and human verification difficult.
- Journals, community Lean checks and the Clay Mathematics Institute have begun months‑long reviews to adjudicate which claims, if any, hold and the episode has raised urgent questions about attribution, ethics and corporate governance in frontier mathematics.