Question 910
In a Google Cloud security design, the requirement is: A document-search agent should not obtain permissions to delete Cloud Storage buckets merely because it can search them. Which action best satisfies this requirement?
Explanation
Correct Answer
Agent tool permission boundaries addresses this case by the following supported operation: Grant only the tool-specific actions and protected service permissions needed, adding controlled review for irreversible operations. For this scenario (a document-search agent should not obtain permissions to delete cloud storage buckets merely because it can search them), verify that the search agent retrieves permitted documents but deletion tool calls fail authorization.
Incorrect Answers
Model Armor prompt injection screening has a distinct purpose: Configure a supported Model Armor prompt-injection template in blocking mode and keep untrusted retrieved text separate from privileged instructions. It instead applies when a retrieval-augmented agent may ingest instructions hidden in external documents that attempt to override its system rules, not the current requirement (a document-search agent should not obtain permissions to delete cloud storage buckets merely because it can search them). Here, verify that the search agent retrieves permitted documents but deletion tool calls fail authorization.
AI artifact supply-chain integrity has a distinct purpose: Scan build artifacts and enforce supported image provenance or Binary Authorization policy on eligible runtime deployments. It instead applies when a model deployment pipeline consumes third-party container images that might contain known vulnerabilities, not the current requirement (a document-search agent should not obtain permissions to delete cloud storage buckets merely because it can search them). Here, verify that the search agent retrieves permitted documents but deletion tool calls fail authorization.
AI training dataset access minimization has a distinct purpose: Limit training service accounts to designated sanitized data resources with scoped storage and dataset grants. It instead applies when a model-training job should use a pseudonymized dataset rather than a raw customer identity table, not the current requirement (a document-search agent should not obtain permissions to delete cloud storage buckets merely because it can search them). Here, verify that the search agent retrieves permitted documents but deletion tool calls fail authorization.






Professional Cloud Security Engineer Practice Tests