IT Home reports OpenAI has deployed a “control room” system internally to stop risky AI tasks

According to IT Home, OpenAI says it has deployed a “control room” system inside its research facilities to halt potentially risky AI tasks. The setup routes workloads such as model training, evaluation and data processing through a central service that decides whether a job may keep generating output. Researchers must request a unique task ID, which the inference engine checks with the control room before running. OpenAI said the longest delay between a stop order and blocking further requests is 35 seconds, and it is working to reduce that to zero.