Amazon AWS Certified Generative AI Developer - Professional AIP-C01 Practice Test 23

Question 220

Multiple business systems need the same GenAI endpoint. Which integration architecture change best addresses this failure?

Give every downstream app a different unversioned Bedrock wrapper with incompatible response formats.
Put model API credentials into each client package to avoid a centralized service.
Publish a central API contract and shared authorization/observability gateway.
Route all callers through an unauthenticated endpoint because the underlying model is read-only.

Explanation

Correct Answer

Publish a versioned shared API contract and caller-aware gateway so multiple apps reuse a consistent authenticated model service.

Incorrect Answers

Divergent contracts cause integration breakage and inconsistent policy enforcement.

Distributed client secrets expand access exposure and cannot enforce a common boundary.

Prompts and model usage still need identity, quotas and data-access authorization.

Question 221

An enterprise user may access records in only one business unit, while the GenAI service calls several retrieval and update APIs. Select TWO prerequisites for establishing verified identity and enforcing its specific downstream entitlements.

Authenticate enterprise caller identity.
Authorize all employees to use the shared service-role's entire data set after login.
Apply caller-entitled resource permissions in both tool execution and retrieval.
Let the model filter other units' documents only after loading them into the prompt.
Use a cached employee department field forever without checking current entitlements on sensitive actions.

Explanation

Correct Answer

Verify the real enterprise caller identity using the established SSO/trust path.

Enforce that caller's business-unit scope at both tool/API action and retrieval, not merely at sign-in.

Incorrect Answers

Authentication is not unrestricted authorization to other business units.

Restricted content is disclosed during context assembly.

Entitlements can change; action-time authorization must remain authoritative.

Question 222

An agent asks for employee payroll information under a service admin role. What should the engineer implement at the enterprise service boundary?

Enforce effective employee row permissions before delivering model context.
Let the service role read every salary record and tell the model to omit unauthorized results.
Treat an authenticated requester as permitted to read all payroll data in the company.
Return all payroll hits but remove employee names from final citations.

Explanation

Correct Answer

Enforce the employee's actual payroll row permissions server-side before any search result becomes model context, even if the backend service has a broad role.

Incorrect Answers

Once retrieved into the model, private payroll data have already crossed the access boundary.

Identity verification does not imply row-level entitlement.

Redaction of output may still leak financial amounts and does not secure retrieval.

Question 223

A workflow has both personal data and public policy documents. Which integration architecture change best addresses this failure?

Keep their storage, access and permitted-purpose boundaries distinct.
Merge public and private documents into one unrestricted embedding index for simpler search.
Use public-source license as authorization to reuse all related personal data.
Prompt the model not to reveal private content after loading both data sets into its context.

Explanation

Correct Answer

Segregate personal and public policy sources with per-purpose access, encryption and consent before combining evidence for an answer.

Incorrect Answers

Shared access policy can disclose personal data to public-policy callers.

Licensing of policy text does not grant personal-data processing consent.

Instructions cannot substitute for storage and retrieval access boundaries.

Question 224

An integration path crosses AWS and on-premises systems. Which change satisfies the specified security or reliability requirement?

Send confidential on-premise records through arbitrary public endpoints because the model is in AWS.
Assume Direct Connect by itself authorizes every AWS service to access on-premise databases.
Select a cross-Region global inference profile without examining destination geographies.
Use supported private hybrid connectivity and region-compliant data routing.

Explanation

Correct Answer

Choose supported private hybrid connectivity (VPN/Direct Connect etc.) with authenticated routing and verify permitted data processing Regions.

Incorrect Answers

Public path choice can violate security and residency constraints.

Network connectivity does not grant application data permissions.

Hybrid network privacy does not ensure where the model processes prompts.

Question 225

An on-prem database cannot expose public endpoints. What should the engineer implement at the enterprise service boundary?

Open the database to the Internet temporarily so the agent can fetch rows.
Put the database root password into a system prompt and keep the port private.
Apply authenticated private network route/proxy and least-privilege credentials.
Create a public API with no caller validation and rely on TLS alone.

Explanation

Correct Answer

Expose a scoped private proxy/route with workload authentication, database roles and supported hybrid network paths instead of a public database listener.

Incorrect Answers

Temporary public exposure violates the private-endpoint requirement.

Prompt secrets may leak and root access is not least privilege.

Encryption in transit does not establish who may query private records.

Question 226

An application wants to use AWS Outposts for residency. Which integration architecture change best addresses this failure?

Assume every regional Bedrock model endpoint physically runs on the organization's Outposts rack.
Use a local S3 cache to claim that cloud-hosted model inference never leaves the premises.
Verify supported services and actual processing jurisdiction before selecting placement.
Ignore service/Region availability because corporate network connectivity is private.

Explanation

Correct Answer

Verify that the required inference service is actually offered on the chosen Outposts footprint, including real prompt processing and storage jurisdictions.

Incorrect Answers

Regional control plane and inference placement are not automatically local to an Outposts rack.

Storage placement does not determine compute processing location.

Connectivity alone does not prove residency or product support.

Question 227

A company permits sensitive customer prompts to be processed only in approved jurisdictions, but the GenAI service inventory has several model-routing profiles with undocumented destination Regions. Which deployment-wide control must security require?

Keep source documents in a compliant S3 bucket but route prompts through an unrestricted global model profile.
Treat an AWS account's home Region as the model's only possible inference destination.
Permit model calls based solely on provider brand without checking the actual model/Region routing.
Inventory each profile’s destination Regions and pin routing to the permitted processing geography before sensitive prompts are sent.

Explanation

Correct Answer

Data residency depends on actual inference processing destinations, not only source-bucket location or caller Region. A governance control must examine routing paths across the service inventory.

Incorrect Answers

S3 location does not constrain where a model processes prompt content.

Cross-Region inference profiles may route to other listed Regions.

Provider identity alone cannot satisfy specific geographic processing requirements.

Question 228

A monolith needs a new GenAI feature. What should the engineer implement at the enterprise service boundary?

Link the model SDK directly into every monolith route and redeploy all features for each prompt change.
Introduce an independently scalable, versioned API rather than coupling all app deployment.
Expose model admin keys to the browser so the monolith does not need backend changes.
Share one untyped response string between accounting, HR and policy features.

Explanation

Correct Answer

Extract the GenAI function behind a versioned API with independent scaling, auth and release boundaries so the monolith can integrate safely.

Incorrect Answers

Tight coupling increases blast radius and prevents independent governance/scale.

Client-side secrets bypass the required protected integration boundary.

Without a stable contract, changes can break consumers and privilege separation.

Question 229

A webhook sends duplicate delivery events. Which integration architecture change best addresses this failure?

Use stable event ID, deduplication and idempotent downstream processing.
Process each delivery as a new financial transaction even if the event ID repeats.
Use event arrival timestamps as unique IDs without preserving the source event identity.
Trust the model to spot duplicate messages by their prose content.

Explanation

Correct Answer

Deduplicate using the webhook's stable delivery/event ID and ensure downstream writes are idempotent when providers redeliver.

Incorrect Answers

At-least-once delivery can duplicate records or charges.

Retries may arrive later and get new timestamps for the same action.

Semantic text matching cannot guarantee transactional deduplication.

Question 230

A private GenAI API lacks request authentication. Which change satisfies the specified security or reliability requirement?

Leave the endpoint public because Bedrock later checks IAM on the gateway's service role.
Store shared bearer tokens in browser JavaScript and skip expiry or audience validation.
Only filter model output for policy violations after processing anonymous private requests.
Enforce API Gateway or service-side authentication and caller-specific authorization.

Explanation

Correct Answer

Enforce authenticated invocation and resource/tenant-scoped authorization at API Gateway or backend before calling the FM or tools.

Incorrect Answers

Service IAM does not identify or authorize the original human caller.

Exposed tokens can be replayed by unauthorized clients.

Content moderation is separate from request authentication and data rights.

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

Question 220

Explanation

Correct Answer

Incorrect Answers

Question 221

Explanation

Correct Answer

Incorrect Answers

Question 222

Explanation

Correct Answer

Incorrect Answers

Question 223

Explanation

Correct Answer

Incorrect Answers

Question 224

Explanation

Correct Answer

Incorrect Answers

Question 225

Explanation

Correct Answer

Incorrect Answers

Question 226

Explanation

Correct Answer

Incorrect Answers

Question 227

Explanation

Correct Answer

Incorrect Answers

Question 228

Explanation

Correct Answer

Incorrect Answers

Question 229

Explanation

Correct Answer

Incorrect Answers

Question 230

Explanation

Correct Answer

Incorrect Answers

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