I thought we could fix this using Settings > Personalization in both Openai , Claude etc. Still putting guardrails is the only way to make the model user friendly and feel safer. Otherwise god knows what these tools would do.