Content Safety and Moderation

1. What are the four harm categories classified by Azure AI Content Safety?

Hate, Violence, Self-harm, Sexual Hate, Violence, Bullying, Self-harm Violence, Sexual, Harassment, Hate Hate, Sexual, Self-harm, Terrorism

2. What is the purpose of input moderation in content safety?

To block harmful outputs from the system To screen user inputs before they reach the LLM To provide feedback to users on their inputs To manage organization-specific blocklists

3. Why is a safety check in middleware designed to fail closed?

To maximize system performance To ensure no unsafe content is shown to users To allow greater flexibility in outputs To minimize user complaints about moderation

4. How do blocklists and ML classification complement each other in moderation systems?

Blocklists are slower to update than ML classification Blocklists are more reliable than ML classification Blocklists are deterministic while ML classification is probabilistic Blocklists are used exclusively for user-generated content

Continue: Evaluating AI Systems → Review this lesson Retake quiz