AIF-C01 Practice Question 67: Policy Consistency Verification
AWS Certified AI Practitioner (AIF-C01) · Test 7 · question 7 of 10
An insurer's assistant answers employee questions about a written benefits policy in which eligibility turns on a dozen interacting conditions: tenure, contracted hours, dependants and plan tier. Content filters and a denied topics policy are already configured. The residual risk that worries compliance is an answer that sounds reasonable but quietly contradicts the written policy, and they want a mathematically verifiable check that a given answer is consistent with it. Which Amazon Bedrock Guardrails capability provides that?
Show answer and explanation
Answer:
Automated Reasoning checks *use mathematical techniques to validate natural language content against policies you define*. You upload the source document, Amazon Bedrock extracts formal logic rules and a variable schema from it, and at runtime the check can *detect factually incorrect statements in LLM responses by mathematically proving that generated content contradicts your policy rules*, and *provide mathematically verifiable explanations* citing the specific rules involved. AWS names this scenario directly, listing complex rule sets such as mortgage approvals, zoning laws, insurance eligibility, or employee benefits among the cases where the feature is most valuable. Note that the checks run in *detect mode*: they return findings and feedback that your application acts on, rather than blocking (option A).
- B is wrong because a contextual grounding filter scores whether a response is supported by the retrieved passages. It is a valuable defence against unsupported claims, but it is a similarity and entailment judgement; it cannot reason through a dozen interacting conditions and *prove* that a stated conclusion follows from the rules. - C is wrong because denied topics would stop the assistant discussing eligibility at all. That removes the risk by removing the product; the insurer wants correct answers, not no answers. - D is wrong because redacting personal data protects privacy. It does nothing about whether the eligibility conclusion in the answer is right, which is the entire concern here.
Reference: What are Automated Reasoning checks in Amazon Bedrock Guardrails?