Question 20
During operational handoff for this problem: A department wants sentiment analysis in SQL but proposes shipping every customer comment to an ungoverned third-party endpoint. Which guardrail should be included?
Explanation
Correct Answer
Classify data and review function availability and charges before rollout. This reduces recurrence of: Supported Cortex AI functions can provide managed model operations through Snowflake SQL interfaces. The verification should demonstrate: A small approved comment sample yields results while sensitive rows remain inside the approved processing boundary.
Incorrect Answers
Test generated SQL and model semantics against finance-approved calculations. That safeguard is aimed at ANOTHER risk: A business team asks an LLM to answer revenue questions but the model frequently invents column names and joins. It is less targeted than the control for: Supported Cortex AI functions can provide managed model operations through Snowflake SQL interfaces.
Pin availability checks to target regions in AI application deployment reviews. That safeguard is aimed at ANOTHER risk: A team treats Snowflake's managed AI inference as having exactly the same regional availability for every model. It is less targeted than the control for: Supported Cortex AI functions can provide managed model operations through Snowflake SQL interfaces.
Review min and max cluster settings against real hourly demand. That safeguard is aimed at ANOTHER risk: A multi-cluster warehouse has a large fixed minimum cluster count during a quiet overnight period. It is less targeted than the control for: Supported Cortex AI functions can provide managed model operations through Snowflake SQL interfaces.