guardrails: Policies and restrictions placed on AI models to ensure data is handled responsibly and that the model doesn't create disturbing content. hallucination: An incorrect response from AI.