Amazon AWS Certified Generative AI Developer - Professional AIP-C01 Practice Test 34

Question 330

An application uses an IAM user secret baked into ECS image. Which security or privacy control closes the described gap?

Replace with scoped task role or workload credentials, revoke exposed key.
Rotate the hardcoded secret only when the container image is rebuilt annually.
Set the environment variable name to an obscure string without removing the secret.
Give every ECS task a common AdministratorAccess policy instead of a user key.

Explanation

Correct Answer

Use an ECS task role or equivalent workload identity with scoped permissions; revoke the exposed static IAM user secret embedded in the image.

Incorrect Answers

An exposed long-lived key remains a compromise risk.

Obfuscation cannot secure credentials in deployed artifacts.

Workload identity is useful only with least-privilege resource permissions.

Question 331

A Lambda action is permitted to delete unrelated S3 buckets. Which control is required in this deployment?

Leave s3:DeleteBucket on Resource * so the tool can handle any future customer.
Put a natural-language prompt warning around the Lambda function but keep wildcard IAM rights.
Use a bucket encryption key as the only control over destructive operations.
Reduce its execution role and validate resource-based action policies.

Explanation

Correct Answer

Reduce the Lambda role's S3 delete permissions to explicitly authorized buckets/objects, checking resource policies and action trust.

Incorrect Answers

This enables deletion of unrelated tenant storage.

Prompt guidance cannot prevent unauthorized SDK calls.

Encryption does not enforce which buckets the role can delete.

Question 332

An organization needs to discover sensitive data in S3 source sets. What should be verified for this identity, network or data boundary?

Use a Bedrock model to guess which S3 objects contain PII without scanning source objects.
Enable CloudWatch metrics on a bucket and treat object counts as sensitive-data classification.
Publish Macie findings openly to all users instead of controlled security responders.
Use Macie where applicable to classify sensitive S3 data and route findings.

Explanation

Correct Answer

Amazon Macie detects/classifies sensitive objects in S3 and produces actionable findings; it does not independently authorize all downstream usage.

Incorrect Answers

A generic FM summary is not a validated S3 sensitive-data discovery job.

Storage counts do not identify PII categories.

Findings can reveal sensitive locations and require access control.

Question 333

An agent copies PII into third-party MCP tool arguments. Which security or privacy control closes the described gap?

Send entire customer profiles to every tool to ensure the model has enough context.
Use an MCP tool's advertised description as consent for third-party data sharing.
Apply purpose-based field filtering and outbound authorization before tool call.
Allow the model to add unvalidated private fields to outbound arguments.

Explanation

Correct Answer

Before outbound MCP tool calls, minimize parameters by permitted purpose and check actor-bound authorization for the external target.

Incorrect Answers

Excess attributes leak personal data beyond tool need.

Tool metadata cannot grant data-subject or enterprise permissions.

Model-created arguments are not a data-loss prevention or policy boundary.

Question 334

A support conversation includes account numbers in model context. Which control is required in this deployment?

Drop arbitrary four-digit substrings and assume all remaining numbers are safe.
Preserve unmasked account numbers inside system prompts for future convenience.
Use detected sensitive spans and safe anonymization preserving task semantics.
Rely on model prose telling the user not to share data after the data has been logged.

Explanation

Correct Answer

Identify sensitive account spans and mask/tokenize appropriately while retaining only authorized information the task truly needs.

Incorrect Answers

PII detection requires context and correct entity boundaries.

A system prompt is not a secure personal-data vault.

A warning cannot undo exposure through prompt/log retention.

Question 335

An enterprise inference service must trace failures while avoiding disclosure of private source documents through Bedrock invocation logs. Which TWO requirements should guide its application and logging design?

Enable full raw prompt and document logging for every user and allow broad organization read access to the logs.
Assume Bedrock invocation logging captures every endpoint, all tool calls and all Knowledge Base retrieval operations automatically.
Retain pseudonymous request IDs and only the necessary model, timing, status and token metadata for correlation.
Avoid storing sensitive request/response bodies unless explicitly needed, and enforce scoped log access and retention.
Publish logging buckets publicly so response troubleshooting does not require IAM review.

Explanation

Correct Answer

Correlation identifiers enable diagnostics across stages without routinely copying documents or personal content into observability systems.

Limit sensitive payload collection, retention and visibility according to the data classification and approved investigative need.

Incorrect Answers

Raw prompts can contain secrets and PII. Broad logging access violates the objective of protecting private documents.

AWS invocation logging is disabled by default and has endpoint/operation limits, notably bedrock-runtime; application tools and retrieval need separate telemetry.

Public logging storage is incompatible with private enterprise information and provides no meaningful trace integrity benefit.

Question 336

A cross-account data source uses a broad bucket resource policy. Which security or privacy control closes the described gap?

Use Principal * with s3:GetObject on the whole bucket because both accounts belong to AWS.
Disable versioning to hide previous copies rather than fix broad resource policy.
Assume caller-side IAM Allow always overrides a bucket resource Deny.
Grant only approved principals and required S3 access paths.

Explanation

Correct Answer

Limit bucket policies to named trusted principals, specific objects/actions and account conditions required by the cross-account source workflow.

Incorrect Answers

AWS account membership is not data-owner authorization.

Versioning does not grant or revoke caller permissions.

Effective cross-account permissions require compatible resource and principal policies.

Question 337

An app shares reusable vector embeddings across departments. Which control is required in this deployment?

Assume vectors contain no sensitive information because they are numeric arrays.
Pool all departments under one unrestricted retrieval role and ask the FM to conceal wrong hits.
Check if embedding/metadata can reveal restricted source content and isolate tenants.
Encrypt the common index but allow every tenant to issue unrestricted queries.

Explanation

Correct Answer

Treat embeddings and metadata as potentially revealing source information; isolate/authorize by department before vector search results are shared.

Incorrect Answers

Embeddings and associated metadata may still leak source membership or semantic content.

Retrieval itself exposes unauthorized context.

Encryption cannot enforce per-tenant result authorization.

Question 338

A tool receives a document it is not allowed to read. What should be verified for this identity, network or data boundary?

Return a protected document because the agent's IAM service role technically allows GetObject.
Hide the object location while giving the model the document text.
Trust a username inside the tool request body without verifying its signed identity.
Enforce source permissions for actual caller even when agent has backend service role.

Explanation

Correct Answer

Enforce effective source document permissions using the actual authenticated caller, even when the agent's backend service role can fetch the object.

Incorrect Answers

Backend capability does not imply caller entitlement.

Suppressing the URL does not prevent data exposure.

An untrusted string cannot authorize private document retrieval.

Question 339

A model response is safe text but tool can move money. Which security or privacy control closes the described gap?

Require independently authenticated payment authorization and amount limits.
Treat a safe-content guardrail result as authorization to move funds.
Execute payment tool calls under the agent's AdministratorAccess role for every user.
Use model confidence over 0.9 as evidence the user approved the charge.

Explanation

Correct Answer

Require independent business payment authorization with actor identity, record and amount limits before executing a tool, regardless of output-text safety.

Incorrect Answers

Text moderation and financial permission are separate controls.

The executor must not bypass account/amount entitlements.

A model score is not authenticated consent or transaction approval.

Question 340

A regional model logs abuse-detection data outside permitted scope. Which control is required in this deployment?

Assume prompts never leave the calling Region because the S3 source bucket is local.
Rely on the model provider's brand as a substitute for reviewing current retention policies.
Verify provider-specific storage and cross-Region processing terms before use.
Route through a global profile and inspect only the initiating endpoint's Region.

Explanation

Correct Answer

Inspect exact provider/model/Region-specific input/output abuse-detection storage and cross-Region terms before processing restricted content.

Incorrect Answers

Cross-Region inference and monitoring destinations may differ from source storage.

Different products and versions may have distinct processing/storage terms.

Actual destination Regions and abuse monitoring scope determine exposure.

Pass Your Amazon AWS Certified Generative AI Developer - Professional AIP-C01 Exam Easy!

Question 330

Explanation

Correct Answer

Incorrect Answers

Question 331

Explanation

Correct Answer

Incorrect Answers

Question 332

Explanation

Correct Answer

Incorrect Answers

Question 333

Explanation

Correct Answer

Incorrect Answers

Question 334

Explanation

Correct Answer

Incorrect Answers

Question 335

Explanation

Correct Answer

Incorrect Answers

Question 336

Explanation

Correct Answer

Incorrect Answers

Question 337

Explanation

Correct Answer

Incorrect Answers

Question 338

Explanation

Correct Answer

Incorrect Answers

Question 339

Explanation

Correct Answer

Incorrect Answers

Question 340

Explanation

Correct Answer

Incorrect Answers

AWS Certified Generative AI Developer - Professional AIP-C01 Practice Tests