OpenAI said it is tightening limits on the weapons-related information ChatGPT provides to teens, even when a question appears harmless