| Mobile Web

OpenAI builds kill switch to stop out-of-control agents within 35 seconds

OpenAI has built a central control system that can block model outputs in internal AI research work when safety or security concerns are raised. Cointelegraph reported on Nov. 10 that the system, called a “control room,” tracks model tasks across the research infrastructure and can quickly stop output if problems occur. It applies to training experiments, evaluations and data processing, though shutdown can take up to 35 seconds.