OpenAI describes Control Room, a system that can stop its research model jobs

OpenAI's safety newsroom lists a post, Control Room, under 9 November 2026. The post page itself does not print a date. The post says Control Room tracks model workloads across OpenAI's research infrastructure and can stop them when safety or security concerns arise. It says it connects every research inference workload, including agentic sampling for model training and evaluation and data processing. It says model output halts within 35 seconds of a decision to stop or pause a job, and that the lag comes from a caching layer. It says OpenAI is working to drive that number down to zero.

The post says every research workload that does agentic inference must be associated with an approved safety case. It says inference engines ask Control Room for permission before fulfilling a request, and that a stopped job is refused. It says the system fails closed: if Control Room goes down or is disabled, including by tampering, inference requests are denied until it resumes. It says fail-closed is rolled out now to all frontier model research workloads, and will finish rolling out across all remaining research workloads over the next few days. It says all frontier reinforcement-learning jobs run with a monitor, another model that can tell Control Room to stop a job. It says the external model-serving stack already has a way to take down model deployments. The post does not state a price or a new model name.

What it means for business: this describes a control on OpenAI's own research jobs, not a new product or a customer setting. The 35-second figure and the fail-closed rollout are OpenAI's account of its research infrastructure. This describes the post. It is not advice on a deployment.