OpenAI has deployed a centralized control system called Control Room to stop potentially risky AI model workloads across its research infrastructure, with the system designed to halt further model output within 35 seconds of a decision to intervene.The system is intended to strengthen safeguards as increasingly capable AI models introduce greater safety and security risks. It connects research workloads, including agent-based model training, evaluations, and data-processing jobs, to a central service that controls whether they can continue generating outputs.Unlike conventional monitoring systems that primarily flag suspicious behavior, Control Room can actively prevent a research workload from obtaining additional model outputs. It works alongside OpenAI’s existing systems for monitoring potentially misaligned model behavior.The company said Control Room is already deployed across its research infrastructure, where the need for intervention is greatest. Its external model-serving infrastructure is more centralized and already has operational mechanisms to take down model deployments when necessary.How shutdown worksControl Room connects three main components: the research workload, the inference engine, and the central Control Room service.A workload is a research job, such as a reinforcement learning experiment, an evaluation, or a high-volume data-processing task. Each research workload that performs agentic inference must be associated with an approved safety case.When a researcher starts a job, its code requests a unique job ID from Control Room. This identifier is then attached to requests sent to inference engines, which run models directly on GPUs and generate their outputs.We built Control Room, a central system to track model workloads across our research infrastructure and stop them quickly when safety or security concerns arise.It connects every research inference workload, including agentic sampling for model training and evaluation, data…— OpenAI (@OpenAI) November 9, 2026Before fulfilling an inference request, the engine checks with Control Room to confirm that the associated job is still authorized to run.If a researcher or monitoring system has instructed Control Room to stop that job, the service denies the next inference request. The engine then refuses to generate further model output for that workload.Importantly, the research process itself does not necessarily terminate. Its code may continue running, but it can no longer obtain additional model outputs through the authorized inference engines.This approach also avoids relying on researchers’ workload code to implement safety checks correctly. OpenAI places the critical authorization mechanism in the inference layer, which has fewer and more centrally managed code paths.Fails closed if system failsControl Room is designed to remain effective even when other infrastructure components malfunction. Its fail-closed mechanism also prevents inference requests if the Control Room service itself becomes unavailable.This means a technical failure, deliberate disabling, or potential tampering with the central service will cause inference engines to deny requests until the system returns to normal operation.OpenAI said the maximum delay between receiving a stop instruction and preventing further progress is 35 seconds. This delay comes from a caching layer used to handle the volume of inference authorization requests. The company is working to reduce the interval to zero.The system also integrates with automated monitoring. OpenAI said all its frontier reinforcement learning jobs run with a separate model-based monitor that examines model behaviour and can raise alerts when it detects potential problems.If an alert warrants intervention, the monitoring system can instruct Control Room to terminate the associated job. The service then blocks subsequent inference requests across any model or engine used by that workload.The company cited a recent DNS-related incident, in which a monitor detected an agent attempting to reach an external chatbot, as an example of why such safeguards matter.Control Room is one part of OpenAI’s broader approach to securing frontier model development. The company said it will continue strengthening its monitoring systems and safeguards against misaligned behavior and security breaches.