OpenAI built a centralized control room to track research-model workloads and halt model output within 35 seconds.

OpenAI describes its control room as a central system for tracking model workloads within its research infrastructure. Its stated purpose is to stop those workloads when safety or security concerns arise.

FinancialJuice separately reports that the control room can stop model output in 35 seconds. This is a reported output-stoppage capability, not a disclosed time to power down GPUs, servers or entire data centers.

The supplied messages do not describe the control architecture, testing conditions or number of workloads covered. They also do not establish that the same shutdown capability applies to every OpenAI production service or customer deployment.