OpenAI تكشف Control Room: نظام مركزي يوقف مخرجات نماذجها البحثية خلال 35 ثانية لو ظهر خطرOpenAI unveils Control Room, a central system that halts its research models' output within 35 seconds if a risk appears

OpenAI on 10 November described an internal system called Control Room that tracks every workload running its models across its research infrastructure and can quickly stop any model's output when a safety or security concern comes up. The company says it halts output within 35 seconds of a decision to intervene.

How it works: each research job, such as a training experiment or an evaluation, gets an ID from Control Room, and every workload that does agentic inference must be tied to an approved safety case. Before a model answers any request, the inference engine asks Control Room whether the job is still allowed to run. If a person or a monitoring system has stopped it, the engine refuses any new request, even if the job's own code is still running.

OpenAI says all of its frontier reinforcement learning jobs run with a monitor, another model that inspects everything the model being trained does and raises an alert if it finds a problem. The system is designed to fail closed: if Control Room goes down, or someone deliberately disables it, the engines deny all requests until it is back.

The company says the 35-second lag comes from a caching layer for inference authorisation and that it is working to bring it down to zero. The system covers research infrastructure; for the models people use, OpenAI says it already has a well-developed ability to take deployments down quickly. Fail-closed is live for frontier model research workloads and will reach the remaining research workloads over the next few days. All of this comes from OpenAI itself, with no independent review.