Feed
OpenAI Launches Control Room With 35-Second Kill Switch For AI
AI

OpenAI Launches Control Room With 35-Second Kill Switch For AI

10 Nov 2026, 11:39 am · 10d ago · 1 min read · Inc42

OpenAI has deployed Control Room, a centralised monitoring framework designed to oversee AI research infrastructure and stop rogue models within 35 seconds of an intervention decision. Advanced training jobs are continuously monitored by a separate AI system capable of flagging problematic outputs. When issues arise, Control Room terminates the job and blocks inference requests across all associated engines. While caching layers currently cause a 35-second shutdown delay, OpenAI aims to reduce this latency to zero. The system safeguards model training, evaluations, agentic sampling, and large-scale data processing workflows across research clusters.