Amazon AWS Certified Generative AI Developer - Professional AIP-C01 Practice Test 18

Question 170

A supervisor chooses between code, search and finance specialists. Which change preserves actual tool authorization and trustworthy state?

Allow the supervisor to grant finance privileges just by naming the finance specialist.
Send every task to every specialist, then run all suggested actions in parallel.
Route by bounded capability contracts and authorize each specialist separately.
Let a model select a privileged specialist based on untrusted retrieved role descriptions.

Explanation

Correct Answer

Route specialist calls by documented capability and require independent resource authorization at each specialist/tool boundary.

Incorrect Answers

Tool routing does not create employee financial permissions.

Unnecessary action paths expand costs and unauthorized side-effect risk.

Retrieved text is not an authoritative capabilities or identity registry.

Question 171

An agent acts using one shared admin token for all employees. What should the developer implement before allowing the agent to act?

Use delegated/actor-bound tool access and record the verified initiator.
Give the agent a single global administrator secret in its system prompt.
Log only the agent service name while executing as admin for every employee.
Trust the username the model writes into a tool argument without authentication.

Explanation

Correct Answer

Use actor-bound delegated credentials and audit the real initiating principal; a shared administrator token hides who approved each action.

Incorrect Answers

A prompt credential is exposed to prompt injection and lacks per-user authorization.

Service identity alone cannot prove who requested the specific operation.

Model-generated identity can be spoofed and cannot grant delegated rights.

Question 172

A company exposes internal REST endpoints as MCP tools through AgentCore Gateway. An authenticated caller must not automatically inherit permission to every backend API. What configuration is required?

Leave the gateway unauthenticated and grant its tool target blanket access to production systems.
Define typed Gateway targets and enforce appropriate inbound caller authorization plus separately scoped outbound target authorization.
Place a permission policy in the model's system prompt and permit all caller tool requests at the gateway.
Use a single shared administrator token for both client access and every downstream API.

Explanation

Correct Answer

Gateway schemas expose the capabilities, inbound authorization gates the caller, and outbound authorization controls how the target service is invoked.

Incorrect Answers

An unauthenticated gateway to sensitive targets allows arbitrary callers and cannot establish the true actor for a tool operation.

Model instructions do not enforce the gateway's IAM/JWT policies or the target's business authorization.

Shared administrator credentials collapse least privilege and fail to preserve caller-specific rights.

Question 173

An AgentCore Gateway accepts a valid caller JWT, but a protected backend API requires an OAuth token for that specific employee. What distinction must the implementation preserve?

Use the shared gateway administrator identity as the backend user, because inbound JWT validation is already sufficient for per-user authorization.
Skip the gateway caller check and assume a successful downstream OAuth token acquisition proves the requester was authorized.
Validate the inbound caller, then acquire or exchange separately scoped outbound credentials for the backend target on behalf of the correct user.
Send the JWT claims to the foundation model and let it decide whether to reveal the employee's access token.

Explanation

Correct Answer

Inbound authentication and target-specific outbound authorization solve different boundaries; use supported OAuth/on-behalf-of mechanisms where applicable.

Incorrect Answers

Authentication at the entry point does not grant all backend resource rights; the target must see an appropriate delegated identity or policy decision.

An outbound token alone does not establish who invoked the gateway or whether that caller is permitted to use the tool.

Bearer tokens should be handled by trusted authorization components, not exposed in model context or generated answers.

Question 174

An MCP tool schema was changed to require additional parameters; older agent calls now fail validation. Select TWO client-contract changes needed to reestablish compatible calls, excluding unrelated cache and trust settings.

Version and validate the updated MCP tool schema.
Refresh client-side required arguments before retrying.
Send the old request structure and suppress validation errors until the tool accepts it.
Expose the entire backend API under an administrative wildcard rather than updating the tool contract.
Let the LLM invent defaults for new mandatory fields without verifying intended values.

Explanation

Correct Answer

Version and validate the changed MCP tool schema before invoking it.

Refresh required client arguments to match that deployed tool contract, testing both success and typed errors.

Incorrect Answers

Ignoring schema mismatch cannot create missing required fields.

Broader permissions do not fix versioned parameter compatibility.

An invented required value may change the target business operation.

Question 175

A Lambda tool can run longer than the caller deadline. Which agent/tool architecture control directly resolves the situation?

Assume a timeout response means the Lambda made no database changes.
Tell the user the operation succeeded because the model already generated a confirmation.
Increase the frontend timeout without examining API Gateway and Lambda service limits.
Enforce timeouts and a truthful pending/unknown status instead of invented success.

Explanation

Correct Answer

Bound the Lambda execution against the caller's remaining deadline and return PENDING/UNKNOWN when no authoritative completion receipt exists.

Incorrect Answers

A timeout can occur after a side effect committed.

The confirmation is not evidence from the timed-out backend.

An unsupported timeout setting cannot make a longer backend transaction safe.

Question 176

An action tool fails after partial database writes. Which change preserves actual tool authorization and trustworthy state?

Provide compensating transaction or reconciliation logic with business idempotency.
Blindly repeat the full transaction with a new operation ID after the first partial write.
Delete the trace and report that the whole operation rolled back automatically.
Have the model invent a balancing credit without checking the authoritative ledger.

Explanation

Correct Answer

Use downstream idempotency and a verified compensating/reconciliation workflow for partially committed steps rather than assuming full rollback.

Incorrect Answers

This can duplicate already-committed changes.

A distributed partial commit requires explicit transaction evidence and compensation.

Generated compensation amounts are not a controlled business recovery operation.

Question 177

An agent needs to ask missing questions before a booking. What should the developer implement before allowing the agent to act?

Book the cheapest default room without checking the customer's missing dates.
Send an incomplete tool payload and hope the booking API supplies harmless defaults.
Use explicit user-input step and validate each required field before committing.
Ask the model to infer missing credit card authorization from conversation sentiment.

Explanation

Correct Answer

Request missing booking details through an explicit user-input step and verify each against the authenticated actor and tool schema before commit.

Incorrect Answers

Guessed dates or preferences can create an unwanted charge.

Required reservation fields cannot be assumed safe defaults.

Sentiment is not an authorization signal for a financial action.

Question 178

An agent needs to evaluate code in an isolated session. Which agent/tool architecture control directly resolves the situation?

Execute model-generated Python on the production server with its standard service account.
Use governed sandbox/code interpreter rather than production-shell privileges.
Grant shell/root access to make package imports easier during evaluation.
Send source credentials into the code cell as plain-text variables and rely on the model not to print them.

Explanation

Correct Answer

Run untrusted code in an isolated, permissioned sandbox/AgentCore Code Interpreter, not in the application's production execution environment.

Incorrect Answers

Generated code may access private files or systems beyond the intended task.

Root privileges dramatically enlarge blast radius without helping isolation.

Code execution and outputs can expose embedded secrets.

Question 179

An AgentCore agent has excessive filesystem access. Which change preserves actual tool authorization and trustworthy state?

Mount the organization's entire shared filesystem into every agent session.
Give every agent workspace write access to production configuration for troubleshooting.
Rely on the model to avoid opening confidential files while granting those filesystem permissions anyway.
Isolate execution and constrain resource permissions to assigned task.

Explanation

Correct Answer

Constrain agent runtime filesystem and IAM permissions to the task; runtime microVM isolation is not a substitute for resource authorization.

Incorrect Answers

Session isolation does not make overly broad mounts safe.

Task execution does not require unrestricted production mutation privileges.

Model instructions are not access controls.

Question 180

A team uses one long conversation for permanent legal-case storage. What should the developer implement before allowing the agent to act?

Treat the runtime's in-memory transcript as the sole archive of legal-case evidence.
Move durable case records into authorized persistent storage with deletion controls.
Store all cases in one unpartitioned prompt transcript shared by agents.
Copy case evidence to arbitrary external model requests without legal retention or access policy.

Explanation

Correct Answer

Put durable legal-case records into an authorized persistent store with retention/deletion controls; conversation context is not compliant long-term storage.

Incorrect Answers

Ephemeral sessions can expire and do not supply governed record retention.

Shared context destroys per-case confidentiality and audit separation.

Uncontrolled copies defeat purpose, deletion and jurisdiction requirements.

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

Question 170

Explanation

Correct Answer

Incorrect Answers

Question 171

Explanation

Correct Answer

Incorrect Answers

Question 172

Explanation

Correct Answer

Incorrect Answers

Question 173

Explanation

Correct Answer

Incorrect Answers

Question 174

Explanation

Correct Answer

Incorrect Answers

Question 175

Explanation

Correct Answer

Incorrect Answers

Question 176

Explanation

Correct Answer

Incorrect Answers

Question 177

Explanation

Correct Answer

Incorrect Answers

Question 178

Explanation

Correct Answer

Incorrect Answers

Question 179

Explanation

Correct Answer

Incorrect Answers

Question 180

Explanation

Correct Answer

Incorrect Answers

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