Amazon AWS Certified Generative AI Developer - Professional AIP-C01 Practice Test 16

Question 150

A prompt chain combines extraction, verification and reply. Which prompt or application control best resolves the requirement?

Use one untyped free-text chain and infer which stage succeeded from the final prose.
Send all stage outputs straight to the customer before validation.
Use explicit processing stages with typed outputs and error routing.
Continue to answer generation when a required verification stage times out.

Explanation

Correct Answer

Explicit typed extraction, verification and reply stages make wrong intermediate data rejectable and failures traceable before the final message.

Incorrect Answers

Untyped intermediates hide which extraction or validation step failed.

A partial extraction may contain errors or sensitive fields that should be rejected.

A missing verification result is not evidence that the extracted facts are valid.

Question 151

A prompt flow's conditional branch is never reached. Which action preserves the correct instruction or output boundary?

Test branch predicates against representative responses and failures.
Delete the failed branch and always run the same output node.
Increase the model context limit without inspecting the condition expression.
Check only that the flow deploys, never which branch it takes under test inputs.

Explanation

Correct Answer

Test conditional branch predicates with true/false, null, and error fixtures, checking that the expected flow node is actually executed.

Incorrect Answers

Removing a branch does not implement the required conditional behavior.

Token capacity has no direct effect on a misconfigured branch predicate.

Successful deployment cannot demonstrate correct conditional routing.

Question 152

A new prompt leaks secrets via trace logging. Which change is necessary before this prompt flow is promoted?

Save full raw prompts and credentials in a shared log bucket to aid debugging.
Separate prompt telemetry from sensitive values and minimize stored payloads.
Mask secrets only in the UI while retaining the unredacted tracing backend.
Disable all audit correlation identifiers while continuing to store secret-bearing payloads.

Explanation

Correct Answer

Minimize trace payloads, control logger permissions and retention, and verify that prompt parameters never serialize secrets into diagnostics.

Incorrect Answers

Accessible logs can become a secondary secret disclosure path.

The retained trace still contains sensitive content.

Losing correlation IDs does not protect secrets and harms incident investigation.

Question 153

A chatbot has no replayable prompt release evidence. Which prompt or application control best resolves the requirement?

Record only the date a prompt was edited while allowing its version and sources to change.
Store the output text but not the model ID or inference parameters.
Preserve a UI screenshot instead of executable prompt and source configuration.
Record prompt version, model version, parameters and source snapshot IDs.

Explanation

Correct Answer

Tie a replay to immutable prompt version, concrete model and inference settings, and the exact source-document snapshot identifiers.

Incorrect Answers

A timestamp cannot reproduce which instructions and evidence led to an answer.

Different model versions and settings can produce materially different behavior.

A screenshot is not a replayable technical input fixture.

Question 154

A model gives different answers to equivalent prompts. Which action preserves the correct instruction or output boundary?

Compare unrelated user questions and treat all different responses as inconsistency.
Compare semantic consistency using a repeatable controlled evaluation set.
Evaluate identical prompts at changing model versions and sampling settings without recording changes.
Require literal string equality between every paraphrased answer.

Explanation

Correct Answer

A controlled set of paraphrase pairs distinguishes substantive inconsistency from harmless stylistic variation while holding model settings fixed.

Incorrect Answers

Different underlying tasks need not have equivalent answers.

Uncontrolled variables confound the cause of any answer difference.

Semantically correct varied wording should not be treated as an error.

Question 155

A prompt claims complete task success before tool confirmation. Which change is necessary before this prompt flow is promoted?

Require actual tool receipt and grounded final message after commit.
Treat model-generated words such as success as proof of a database commit.
Generate a plausible confirmation number before calling the order API.
Hide tool timeout details and tell the user the requested action definitely succeeded.

Explanation

Correct Answer

Report completion only after the authorized tool has executed and the authoritative system returned a commit/receipt or confirmed persisted state.

Incorrect Answers

Model intent and actual transactional execution are separate events.

A fabricated number cannot prove a committed business record.

A timeout leaves the transaction outcome unknown until checked.

Question 156

In October 2026 a NEW AWS account with no prior Agents Classic activity needs a managed foundation for an agent with enterprise tools. It plans to call CreateAgent for Bedrock Agents Classic. Which path is supported?

Call CreateAgent repeatedly; a new account is automatically allowlisted after the first error.
Use InvokeInlineAgent as a workaround for an account that has never used Agents Classic.
Design the new agent with Amazon Bedrock AgentCore rather than relying on restricted Agents Classic creation.
Create the agent using an IAM administrator role, which bypasses the customer onboarding restriction.

Explanation

Correct Answer

AgentCore is the supported successor for new agent development, with independent runtime, tool connectivity and memory facilities.

Incorrect Answers

Agents Classic new-account restrictions are an eligibility check, not a transient throttle; retries do not grant prior-use eligibility.

For accounts without prior activity, AWS restricts both CreateAgent and InvokeInlineAgent under the maintenance change.

IAM administrator permissions cannot bypass a service-level restriction on new customer access.

Question 157

An existing, allowlisted Bedrock Agents Classic customer must update a current agent action group in October 2026. An engineer assumes all Agents Classic APIs stopped working for everyone. Which statement is correct?

Agents Classic was deleted for all customers on July 30, 2026, so no existing agent can be changed.
All Bedrock foundation-model inference and Knowledge Bases were retired along with Agents Classic.
Previously eligible accounts may continue using existing Agents Classic and action-group APIs; the new-customer restriction is separate.
An existing eligible account automatically receives every new model version in Agents Classic after July 30, 2026.

Explanation

Correct Answer

AWS maintains the existing APIs for allowlisted customers but restricts new account creation methods; migration to AgentCore is recommended, not mandatory immediately.

Incorrect Answers

AWS explicitly keeps existing allowlisted agents operational in maintenance mode.

Core Bedrock model inference, Knowledge Bases and Guardrails continue separately from the Agents Classic onboarding change.

The Agents Classic model catalog is frozen at the maintenance date; ongoing new model support belongs to newer services.

Question 158

A hotel agent is being given permission to create reservations in the hotel API. Select TWO prerequisites that define a typed booking-tool interface and ensure only the verified customer may execute it.

Expose a typed reservation-tool contract.
Expose the booking backend as a free-form string command and let the model invent parameter names.
Use a shared hotel administrator role for bookings from all signed-in customers.
Mark the booking confirmed as soon as the model emits a toolUse request.
Validate arguments and signed-in user permission before execution.

Explanation

Correct Answer

Define the reservation operation as a typed tool whose input schema specifies dates, guests and allowed fields.

Before invoking the tool, validate both its arguments and the authenticated guest's entitlement to the account/reservation.

Incorrect Answers

A text instruction is not a tested API request schema.

Being signed in does not entitle one user to act on every reservation.

Tool intent is not an actual reservation or a confirmed transaction receipt.

Question 159

An agent receives an API timeout after attempting a booking. What should the developer implement before allowing the agent to act?

Immediately submit a new reservation with a fresh request ID and tell the customer the first booking failed.
Check authoritative transaction state and reuse idempotency key before retrying.
Ask the model to infer booking success from its last generated message.
Increase the model's token limit and resend the entire conversation as a new booking command.

Explanation

Correct Answer

After a booking timeout, inspect committed reservation state under the same idempotency key; a second blind write may create duplicate rooms.

Incorrect Answers

The timeout leaves the first commit unknown; a new ID can charge or reserve twice.

Language-model narration cannot replace the reservation system's transaction receipt.

Context length does not determine whether the original booking committed.

Question 160

A retail agent must remember an item during one chat. Which agent/tool architecture control directly resolves the situation?

Store the shopping-cart state in a shared agent-global variable accessible to all sessions.
Write every one-turn shopping item permanently into all customers' long-term memory.
Rely on the model's pretrained weights to remember the item on the next API call.
Keep short-term session state isolated to the current authenticated actor.

Explanation

Correct Answer

Use actor-bound session state for the current chat; short-term context must not bleed between user or session identifiers.

Incorrect Answers

A global variable can mix different customers' items.

Unscoped persistent storage creates leakage and violates data minimization.

Pretraining does not contain a particular authenticated session's newly provided choice.

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

Question 150

Explanation

Correct Answer

Incorrect Answers

Question 151

Explanation

Correct Answer

Incorrect Answers

Question 152

Explanation

Correct Answer

Incorrect Answers

Question 153

Explanation

Correct Answer

Incorrect Answers

Question 154

Explanation

Correct Answer

Incorrect Answers

Question 155

Explanation

Correct Answer

Incorrect Answers

Question 156

Explanation

Correct Answer

Incorrect Answers

Question 157

Explanation

Correct Answer

Incorrect Answers

Question 158

Explanation

Correct Answer

Incorrect Answers

Question 159

Explanation

Correct Answer

Incorrect Answers

Question 160

Explanation

Correct Answer

Incorrect Answers

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