OpenAI deploys control room that halts runaway agents in 35 seconds

On November 10, 2026, IT Home reported that OpenAI announced the deployment of a "control room" system at its internal research facilities, designed to halt potentially risky AI tasks and to shut down runaway AI agents within 35 seconds. According to the report, the system connects loads such as agent training, evaluation, and data processing tasks to a central service, which then decides whether a task may continue producing output. The system is described as distinct from traditional monitoring that flags suspicious behavior: it can proactively block a risky research task from making further requests for output. When a researcher starts a task, they must request a unique task ID from the control room. That identifier is attached to requests sent to the inference engine, and the engine consults the control room to confirm whether the task may run. If a researcher or a monitoring system issues a stop directive, the control room refuses the next inference request, so the inference engine will not produce any model output for that task. OpenAI stated that the maximum latency from receiving a stop directive to preventing further task execution is 35 seconds. The company attributed this latency to the caching layer used for inference authorization requests and said it is working to reduce the delay to 0. The system also supports automated monitoring integrations. OpenAI disclosed that all frontier reinforcement learning tasks are equipped with an independent model monitor that examines model behavior and raises alerts when potential problems are detected.