Question 280
At fictional Ridgeway Supply Chain, a Snowflake data and AI engineering team is conducting a regional AI data processing evaluation while preserving enterprise data boundaries and a reviewable audit trail. The separately composed technical incident is: An internal agent calls a document search tool that intermittently times out, and the agent fabricates a source-based answer rather than acknowledging missing retrieval results. 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: D (UNVERIFIED)
Preliminary individual option explanations — NOT expert validated
Correct Answer
Exercise tool-failure paths, capture tool status in traces and routinely test the no-retrieval abstention policy. Agents must distinguish tool execution failures from successful empty results and must not fill missing evidence with speculation.
Incorrect Answers
This response concerns Unauthorized tenant content in retrieval, which is different from the stated failure. An internal agent calls a document search tool that intermittently times out, and the agent fabricates a source-based answer rather than acknowledging missing retrieval results.
RAG chunk boundaries break policy answers is insufficient to demonstrate this result: Fault-injection tests show bounded retries, traceable failure messages and no source claims after a failed search call.
Batch classification retry cost explosion is not the cause of the reported failure: An internal agent calls a document search tool that intermittently times out, and the agent fabricates a source-based answer rather than acknowledging missing retrieval results.