AI-300 Practice Question 18: Protected Material Evaluator
Microsoft Machine Learning Operations Engineer Associate (AI-300) · Test 2 · question 8 of 10
Before a marketing copy generator ships, the legal team asks for measured evidence that its output does not reproduce copyrighted song lyrics or passages from published books. Which built-in Microsoft Foundry risk and safety evaluator produces that evidence?
Show answer and explanation
Answer:
Microsoft's table of built-in risk and safety evaluators maps builtin.protected_material to copyrighted content, taking query and response as inputs (option D). Run over a test dataset, the safety evaluators report an aggregate defect rate, "the percentage of undesired content detected in the response from your AI system", which is exactly the measured evidence legal is asking for rather than a subjective read of a few samples.
- A is wrong because the indirect attack evaluator looks for jailbreak instructions smuggled into content the system ingests, such as a poisoned document a retriever returns. That is a prompt-injection risk, not a copyright one. - B is wrong because the code vulnerability evaluator inspects generated code for security flaws. This application generates marketing prose. - C is wrong because hate and unfairness measures language that reflects hate toward, or unfair representation of, individuals and social groups. It is a real harm category, but a different one; copied lyrics can be perfectly inoffensive and still be a legal problem.
Reference: Risk and Safety Evaluators for Generative AI - Microsoft Foundry