Question 10
At fictional Northshore Public Library, a Snowflake data and AI engineering team is conducting an AI retrieval and source-citation assurance review with legal, privacy and information-security review for any restricted inputs. The separately composed technical incident is: A fictional insurer's policy assistant drafts confident employee guidance from a language model even when the requested policy has no supporting internal document, and users cannot distinguish sourced text from unsupported assertions. Assume only these documented facts and use safe, authorized production-change procedures; exact current Snowflake feature syntax and region or account availability require current review. Which repeatable assurance or control is most relevant to preventing this situation?
Explanation
Writer-proposed answer: B (UNVERIFIED)
Preliminary individual option explanations — NOT expert validated
Correct Answer
Maintain a regression set of missing-source and conflicting-source queries, review citation coverage and update retrieval permissions whenever the policy repository changes. A language model's fluency does not establish factual grounding; retrieval, source attribution and evidence-based refusal must be evaluated separately.
Incorrect Answers
This response concerns Agent permissions and external tool boundaries, which is different from the stated failure. A fictional insurer's policy assistant drafts confident employee guidance from a language model even when the requested policy has no supporting internal document, and users cannot distinguish sourced text from unsupported assertions.
Cross-region model inference consent and residency is not the cause of the reported failure: A fictional insurer's policy assistant drafts confident employee guidance from a language model even when the requested policy has no supporting internal document, and users cannot distinguish sourced text from unsupported assertions.
Model-selection latency and reliability tradeoffs alone does not establish the needed evidence: Independent test prompts with present, absent and conflicting policies show cited supporting passages for grounded answers and explicit abstention for unsupported claims.