
OpenAI Launches Control Room to Halt Rogue AI Outputs in 35 Seconds
10 Nov 2026, 4:40 pm · 10d ago · 1 min read · YourStory
OpenAI has introduced Control Room, a centralised safety system designed to halt artificial intelligence model outputs flagging security or safety concerns within 35 seconds. Integrated across OpenAI's internal research infrastructure, the feature connects inference workloads, including agentic sampling for model training, evaluation, and high-volume data processing tasks. Control Room operates by requiring inference engines to validate requests with a central service before fulfilling them; if a job is stopped, output generation is denied immediately. The system also runs automated checks to detect unauthorized engine modifications, mitigating risks as autonomous AI agents grow more capable.