Overview
- OpenAI said on Tuesday that an internal system running roughly 10,000 coordinated AI agents worked for about 88 hours to produce a proposed Navier–Stokes 'blowup' proof and a machine‑checkable Lean formalization.
- The company said it will not claim the $1 million Millennium Prize and released the proof publicly so mathematicians can inspect and verify the work.
- NYU mathematician Tristan Buckmaster and Anthropic researcher Levent Alpöge say they had been working on a closely related approach and that Buckmaster’s unpublished drafts were stored in OpenAI’s Codex, prompting accusations that OpenAI rushed or used their work.
- OpenAI denies that its researchers or agents saw the unpublished drafts and says no specific user data was accessed, but it also stated it cannot rule out that de‑identified product‑derived data contributed to model improvements.
- Reporters and researchers flagged that the internal model used an opaque training or reasoning technique that reduces readable chain‑of‑thought, and the episode highlights how concentrated compute and closed‑door agent systems are reshaping credit, reproducibility, and governance in academic research.