A developer uses Azure OpenAI to generate customer support responses. The developer wants to ensure that the model does not produce responses that contain offensive, hateful, or harmful language, even when users input problematic prompts. Which Azure OpenAI feature should the developer configure to achieve this?