OpenAI has introduced Control Room, a central system that tracks model workloads across its research infrastructure and can stop them quickly when safety or security concerns arise. The system connects every research inference workload, including agentic sampling for training and evaluation. It works with existing misalignment monitoring and halts further model output within 35 seconds of a decision to intervene. The delay stems from a caching layer for authorization checks; OpenAI says it is working to reduce that time to zero. The control room is designed to fail closed, so if the service itself is unavailable, all inference requests are denied. Every agentic research job must register with an approved safety case. The system is already deployed across OpenAI’s research infrastructure.
OpenAI has introduced Control Room, a central system that tracks model workloads across its research infrastructure and can stop them quickly when safety or security concerns arise. The system connects every research inference workload, including agentic sampling for training and evaluation. It works with existing misalignment monitoring and halts further model output within 35 seconds of a decision to intervene. The delay stems from a caching layer for authorization checks; OpenAI says it is working to reduce that time to zero. The control room is designed to fail closed, so if the service itself is unavailable, all inference requests are denied. Every agentic research job must register with an approved safety case. The system is already deployed across OpenAI’s research infrastructure.