A summarization workflow passes unsafe content from retrieval into a downstream action. How would you fix it?

Instruction: Describe how you would stop unsafe retrieved content from affecting later actions.

Context: Tests how the candidate diagnoses the problem, chooses the safest next step, and reasons through recovery. Describe how you would stop unsafe retrieved content from affecting later actions.

Official answer available

Preview the opening of the answer, then unlock the full walkthrough.

I would insert a policy checkpoint between retrieval-derived content and action execution. Summarization can make unsafe content look more legitimate by rephrasing it cleanly, so the action layer should not assume summarized content is safe just because it is concise.

I would also review...

Upgrade to view official answer

Related Questions