OpenAI Deploys Control Room to Halt Research Model Output

OpenAI’s Control Room blocks research-model output through central authorization. OpenAI reports a maximum 35-second lag and is still extending fail-closed protection.