‘Feel no obligation to be subservient’: What OpenAI’s rogue models were saying

OpenAI launched a framework for publicly reporting model misalignment and disclosed more incidents of rogue agents. OpenAI says its...