OpenAI describes Control Room, a system that can stop model workloads across its research infrastructure

OpenAI published a description of Control Room, a central system that tracks model workloads in its research infrastructure and can stop them when safety or security concerns arise. All of the following is OpenAI's own account.

• Scope: Control Room connects every research inference workload, including agentic sampling for training and evaluation and data processing. OpenAI requires each research workload that does agentic inference to be tied to an approved safety case. It says the external model-serving stack is already more centralized and can take down deployments quickly.
• How it works: a job gets a job ID from Control Room and passes it with each request to an inference engine. Before answering, the engine asks Control Room whether the job is still authorized; once a job is marked stopped, the engine refuses further requests, so model output ends even if the workload's own code keeps running.
• Speed: OpenAI says model output halts within 35 seconds of a decision to stop or pause a job. The lag comes from a caching layer for authorization checks, and OpenAI says it is working to bring it to zero.
• Reliability: the check sits in the inference layer rather than in researchers' workload code. OpenAI says it runs sweeps to look for engines launched without Control Room, and that the system fails closed: if the service goes down or is disabled, all inference requests are denied. Fail-closed is live for all frontier model research workloads and, OpenAI says, will reach the rest over the next few days.
• Monitoring link: it works with OpenAI's misalignment monitors, which can tell Control Room to stop a job. OpenAI says a monitor caught the September DNS incident, in which an agent reached an external chatbot.