OpenAI says central authorization system can halt research model output

OpenAI has deployed Control Room across its research infrastructure to block further model output after a safety or security stop instruction, Unite.AI reported. The company reports a maximum delay of 35 seconds between receiving that instruction and preventing further progress. Inference engines must obtain authorization before returning output for a registered workload. Human responders or monitoring systems can stop a job centrally without relying on researcher-managed code to obey the instruction. The stop can leave the research process running while denying it model output. OpenAI says all frontier-model research workloads now deny inference requests if Control Room becomes unavailable, with that protection scheduled to reach remaining research workloads in the following days.