# Control Room

OpenAI has deployed Control Room across its research infrastructure, a central system that tracks model inference workloads and stops them quickly when safety or security concerns arise. It covers agentic sampling for training and evaluation, data processing and other research inference jobs, works with misalignment monitoring, and halts model output within 35 seconds of a decision to intervene, a lag that comes from a caching layer for inference authorization. The termination mechanism sits in the inference layer rather than in workload code, and the system denies all inference requests if it goes down.

OpenAI has deployed Control Room across its research infrastructure, a central system that tracks model inference workloads and stops them quickly when safety or security concerns arise. It covers agentic sampling for training and evaluation, data processing and other research inference jobs, works with misalignment monitoring, and halts model output within 35 seconds of a decision to intervene, a lag that comes from a caching layer for inference authorization. The termination mechanism sits in the inference layer rather than in workload code, and the system denies all inference requests if it goes down.