Control Room
OpenAI created Control Room to stop AI models in 35 seconds
The new system allows blocking the generation of results from research models after detecting security threats. It is already operational within OpenAI's internal infrastructure and can stop tasks on command from employees or automatic monitoring. The maximum delay is 35 seconds.