Amazon AWS Certified Generative AI Developer - Professional AIP-C01 Practice Test 15

Question 140

A model generates unrecognized status codes. Which change is necessary before this prompt flow is promoted?

Accept any syntactically valid JSON string as an approved business status.
Require a structured response contract and validate the enum downstream.
Increase generation temperature until one output happens to match an allowed status.
Treat an unspecified fallback string as equivalent to an approved state in downstream code.

Explanation

Correct Answer

Constrain output to the supported schema where available and validate enum values in application code; syntax correctness alone does not ensure business validity.

Incorrect Answers

A valid string may still be an unauthorized enum value.

Random sampling does not enforce a stable contract.

Unknown states must be rejected or handled explicitly, not silently authorized.

Question 141

A versioned Bedrock Prompt Management template inserts a third-party webpage into a content variable. The webpage contains role delimiters that try to override the application’s instruction hierarchy. What control must the prompt assembly preserve?

Copy instructions from the external page into the system role and allow tool execution.
Remove HTML tags but leave embedded instructions with elevated priority.
Let a model determine whether the webpage itself grants access to sensitive tools.
Preserve high-priority application instructions and treat the interpolated third-party webpage text as untrusted prompt data, not privileged roles or tool authority.

Explanation

Correct Answer

A template variable may contain adversarial role syntax but must not become the system/developer instruction layer. Restrict tool permissions and keep the website text in a lower-priority data position.

Incorrect Answers

An attacker could use web content to bypass application policy.

Sanitizing markup does not change the trust boundary of page text.

Authorization is a verified application identity/policy decision, never a page assertion.

Question 142

A support conversation exceeds history length. Which action preserves the correct instruction or output boundary?

Increase model context indefinitely and preserve all chat turns without retention controls.
Discard all conversation state after each message and assume the model can reconstruct it.
Reuse one global summary for all customer sessions to reduce storage.
Summarize durable state into a governed store and truncate context safely.

Explanation

Correct Answer

Store a bounded, governed case summary and useful references, then selectively reload authorized history rather than replaying the entire growing transcript.

Incorrect Answers

Every model has finite context and retaining unneeded sensitive turns enlarges risk and cost.

Missing prior facts can break the support workflow.

Shared unsafely scoped history risks cross-user disclosure.

Question 143

A user asks to resume prior support case in a new session. Which change is necessary before this prompt flow is promoted?

Query every prior support case by approximate name without verifying account ownership.
Rely on the model to remember customer facts when its prior session is not supplied.
Retrieve only their authorized case state from a persistent application store.
Copy all users' conversation memory into a shared new-session prompt.

Explanation

Correct Answer

Load permitted persistent case records under the newly authenticated actor; a new session alone does not inherit arbitrary previous-user history.

Incorrect Answers

Names are not sufficient proof of who may read stored case history.

An independent model invocation does not guarantee access to another session's state.

Global transcript sharing violates least privilege and session isolation.

Question 144

A chat request lacks critical account identifiers. Which prompt or application control best resolves the requirement?

Ask a targeted clarification before calling account-mutating tools.
Guess the most recently used account from model-generated text and execute immediately.
Call the transaction tool with a blank account ID and rely on backend defaults.
Select an unrelated account from a cached support transcript without asking the signed-in user.

Explanation

Correct Answer

A missing account identifier is a required business input; ask the user and authenticate ownership before attempting a mutating tool operation.

Incorrect Answers

A plausible guess is not identity verification and could change the wrong account.

Implicit defaults can target the wrong record or violate required request contracts.

Cached facts cannot substitute for fresh required transaction input.

Question 145

A prompt uses untrusted variables in an executable template. Which action preserves the correct instruction or output boundary?

Interpolate raw user strings into a privileged tool-call template without quoting.
Disable tool parameter checks because the prompt contains an instruction to be careful.
Strip all whitespace from user values rather than checking their grammar and role.
Escape/validate substituted values before assembling tool calls.

Explanation

Correct Answer

Escaped and type-checked prompt variables prevent user-provided strings from injecting template delimiters or tool arguments.

Incorrect Answers

Untrusted delimiters can alter intended fields and command boundaries.

Prompt intentions do not implement typed validation at execution.

Whitespace removal is not a reliable encoding or authorization control.

Question 146

A prompt update silently increases hallucinated policy claims. Which change is necessary before this prompt flow is promoted?

Approve the prompt because its sample answers sound more fluent than the previous draft.
Run only syntax validation on its output without checking source-supported claims.
Run versioned grounding regression tests before promotion.
Change the model temperature and skip the regression corpus.

Explanation

Correct Answer

Compare the updated prompt with a fixed source-grounded regression set; fail promotion when the new version invents unsupported policy claims.

Incorrect Answers

Fluency is not evidence of factual grounding.

Well-formed text can still misstate a policy.

Sampling changes do not establish that new prompt claims are supported.

Question 147

A model's final JSON has valid syntax but wrong business values. Which prompt or application control best resolves the requirement?

Apply application-level schema and business rule validation.
Accept any parseable object regardless of account entitlement or permitted enum.
Move all validation into an English prompt and remove runtime code checks.
Return a success flag after syntactic parsing without verifying referenced business records.

Explanation

Correct Answer

Business validators check allowed value domains and record references after JSON parsing; a valid JSON object can still contradict account rules.

Incorrect Answers

Schema shape alone is not sufficient to authorize the business value.

The model can violate prompt instructions; code must enforce the contract.

A syntactically valid but incorrect object must not be committed.

Question 148

A Bedrock prompt variant targets an unavailable model version. Which action preserves the correct instruction or output boundary?

Validate selected model and prompt compatibility at release time.
Assume every saved prompt revision can invoke any new model ID in every Region.
Approve deployment based only on the historical draft preview output.
Change the user-facing prompt name while leaving an invalid model binding untouched.

Explanation

Correct Answer

The saved prompt's model binding and region must be supported at release time; an archived variant may point to a retired/unavailable model.

Incorrect Answers

Model availability and prompt API configuration are version/Region specific.

A past preview does not prove current model compatibility.

A renamed label cannot make an unsupported inference target available.

Question 149

An FM returns reasoning text instead of requested fields. Which change is necessary before this prompt flow is promoted?

Parse the first brace fragment of the reasoning text as the final business response.
Validate output format and use a bounded correction/exception path.
Return unstructured internal reasoning to the caller as if it were verified JSON.
Retry the model indefinitely with progressively longer instructions until it emits JSON.

Explanation

Correct Answer

Format and schema validation should catch prose in place of required fields; a bounded repair attempt or explicit exception follows, not blind acceptance.

Incorrect Answers

An incidental fragment need not match the required schema or semantic fields.

Reasoning prose is not the requested machine-readable contract.

Unbounded retries are unreliable and can exceed quotas or request budgets.

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.

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

Question 140

Explanation

Correct Answer

Incorrect Answers

Question 141

Explanation

Correct Answer

Incorrect Answers

Question 142

Explanation

Correct Answer

Incorrect Answers

Question 143

Explanation

Correct Answer

Incorrect Answers

Question 144

Explanation

Correct Answer

Incorrect Answers

Question 145

Explanation

Correct Answer

Incorrect Answers

Question 146

Explanation

Correct Answer

Incorrect Answers

Question 147

Explanation

Correct Answer

Incorrect Answers

Question 148

Explanation

Correct Answer

Incorrect Answers

Question 149

Explanation

Correct Answer

Incorrect Answers

Question 150

Explanation

Correct Answer

Incorrect Answers

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