Gives OpenAI a kill switch for rogue model workloads, a safety-operations template for labs.
OpenAI introduced Control Room, a central system that tracks model workloads across its research infrastructure and can rapidly halt them when safety or security concerns arise, reportedly within 35 seconds.