Overview
- OpenAI published a 166‑page paper on Tuesday, September 8 saying an internal ensemble of roughly 10,000 AI agents produced a proposed proof after about 88 hours of work.
- Mathematicians and journals have not yet verified the result, so the correctness and acceptance of the proof as a Clay Millennium Problem solution remain unresolved.
- Tristan Buckmaster and Levent Alpöge have published related work and contend OpenAI knew of their progress before the announcement and offered publication terms that raised authorship concerns.
- OpenAI says its agents found the result independently but acknowledges it cannot fully exclude that anonymized user‑derived data helped improve its models.
- The effort used massive compute and cost estimates in the millions, OpenAI says it will not seek the Clay prize, and the episode raises questions about data governance, credit and how the math community will validate AI‑driven discoveries.