OpenAI said that it has built Control Room, a centralised system to track model workloads and stop them quickly when safety or security concerns arise

The field of preventing harm from advanced AI systems, spanning alignment, evaluation, and decisions to slow or withhold models whose capabilities outpace their safeguards.