OpenAI said on Friday it is tightening limits on the weapons-related information ChatGPT provides to teens, even when a question appears ​harmless, as part of new safeguards aimed at preventing ‌real-world violence.