Overview
- OpenAI says an internal reasoning model found a new family of geometric constructions that outperform the long‑assumed square‑grid arrangement, a claim that was announced in late May 2026.
- Multiple outside mathematicians have checked the construction, it has been formalized in the Lean proof assistant, and a paper has been submitted for journal peer review.
- The model drew on tools from distinct fields, notably algebraic number theory and discrete geometry, to synthesize a solution that human teams had not previously connected.
- Key pieces of evidence have not been released: OpenAI has not published the model's raw chain‑of‑thought or internal reasoning, and experts warn LLMs still hallucinate and require close human verification.
- The episode highlights wider stakes: if AI reliably produces original research it could speed progress in fields from cryptography to biology, yet it also raises calls for transparent data, independent audits, and clearer validation standards.