A malicious document exploits the gap between retrieval and tool use. How would you close it?

Instruction: Explain how you would defend a workflow where retrieved content can influence tools indirectly.

Context: Tests how the candidate diagnoses the problem, chooses the safest next step, and reasons through recovery. Explain how you would defend a workflow where retrieved content can influence tools indirectly.

Official answer available

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

I would put an explicit policy and validation layer between retrieved content and tool execution. Retrieved text should never become tool-driving instructions without being re-evaluated as untrusted material under the action policy.

I would also keep provenance and trust labels attached to retrieved...

Upgrade to view official answer

Related Questions