AIF-C01 Practice Question 20: Hallucination Grounding Check
AWS Certified AI Practitioner (AIF-C01) · Test 2 · question 10 of 10
A bank's question-answering application retrieves passages from its fee schedule and passes them to a foundation model with each customer question. Audits find that some answers still state fees that don't appear in the retrieved passages. The bank wants to automatically block answers that aren't supported by the source passages before customers see them. What should it use?
Show answer and explanation
Answer:
RAG reduces hallucinations, but it doesn't eliminate them. The model can still add details that aren't in the retrieved text. A contextual grounding check in Amazon Bedrock Guardrails takes the reference source, the query, and the model response, then scores two things. Grounding asks whether the response is supported by the source; any new information counts as ungrounded. Relevance asks whether it answers the question. Responses that score below the thresholds you set are detected as hallucinations and blocked. That's automated output validation against the source (option B).
- A is wrong because lower temperature makes output more deterministic, not more truthful. A model can consistently repeat the same invented fee. - C is wrong because a denied topic would block every fee question, which defeats the purpose of the application. It doesn't check whether an answer matches the source. - D is wrong because logging produces an audit trail after the fact. It doesn't stop an unsupported answer from reaching the customer.
Reference: Use contextual grounding check to filter hallucinations in responses