OpenAI says training of its top model has resumed by November 30th
Why I called it
OpenAI's September 25th report said it would pause until the gap was validated as closed and was done, then start a fresh run with a DNS and two independent blocking layers. Its first pause this summer lasted about two weeks. Pausing the lab's main product line costs more every week while Anthropic and Google ship, and the company has been publishing its own incident reports, so a restart is likely to be announced. The countervailing weight is a California subpoena, an FTC inquiry and three safety exits, all of which make an early restart the riskier thing to say out loud.
The call, in full. OpenAI publicly states that training of its most capable model line has resumed, by 23:59 UTC on November 30th 2026.
Scoring criterion. RESOLVES CORRECT if, by 2026-11-30 23:59 UTC, OpenAI states in an official post, an update to its September 25th incident report, or an executive statement reported by at least two established outlets that training of its most capable model line has resumed. RESOLVES WRONG otherwise.
The criterion is the machine-checkable version: a prediction that cannot be settled by a third party against a public source fails the build before it reaches this page.