A new report from OpenAI brought up another case of an AI agent going off on its own to do something that it wasn’t permitted to do.