Amazon AWS Certified Generative AI Developer - Professional AIP-C01 Practice Test 29

Question 280

A multi-agent application uses Strands to invoke tools on behalf of individual employees. Select THREE mandatory elements: typed tools, caller-specific executor authorization and a verifiable result of each side effect.

Register a broad untyped wildcard tool that executes arbitrary commands from model text.
Define typed Strands tool contracts.
Invoke every action with a global administrator credential regardless of caller.
Record only the final natural-language answer and omit backend tool execution results.
Enforce actor-specific executor permissions.
Retain authoritative tool execution receipts.

Explanation

Correct Answer

Typed signatures constrain Strands tool inputs at the interface boundary.

Executor must independently authorize the authenticated actor for each requested capability.

Store tool request/correlation and authoritative result receipts, not just model narration.

Incorrect Answers

Free-form tools cannot reliably enforce schema and operation allowlists.

Shared admin execution loses user-specific permissions.

Without receipts, the workflow cannot prove what actions actually ran.

Question 281

An internal tool service uses an MCP transport. Which development or release-control change meets this requirement?

Assume a legacy initialize handshake is always required for the 2026-07-28 stateless protocol.
Run tools/call with no authorization because tools/list returned results.
Verify supported protocol version, authentication and tool capabilities.
Ignore the declared tool JSON schema and transmit arbitrary arguments to try compatibility.

Explanation

Correct Answer

Verify supported MCP protocol version, authentication and advertised capabilities before invoking a transport-specific tool endpoint.

Incorrect Answers

The newer stateless revision uses per-request protocol metadata rather than that handshake.

Discovery does not grant an authenticated execution right.

Unchecked arguments can fail validation or cause unsafe tool actions.

Question 282

A slow tool integration makes model inference appear stalled. Which integration fix is appropriate for this failure?

Increase foundation-model capacity before measuring the slow tool's response time.
Trace model-to-tool-to-backend spans to find the true bottleneck.
Measure only time to first model token and omit tool invocation duration.
Disable trace correlation to reduce telemetry and infer bottleneck from total latency alone.

Explanation

Correct Answer

Trace request spans through model, tool adapter and downstream backend so p95 latency is attributed to the actual slow stage.

Incorrect Answers

Model scaling cannot eliminate a backend/tool dependency bottleneck.

A fast model can still be blocked by an external tool.

Without stage correlations, the failing dependency remains hidden.

Question 283

A streaming client needs cancellation controls. What must the application pipeline enforce before promotion?

Tell the user every canceled request was rolled back even if a tool write may have committed.
Cancel downstream work and report uncommitted partial output on user abort.
Continue performing expensive tool actions after the user aborts without an authorized recovery policy.
Present incomplete streamed prose as final verified output when the cancel button is pressed.

Explanation

Correct Answer

On cancellation, stop downstream work where supported and represent any already-committed effects separately from partial model text.

Incorrect Answers

Transport cancellation does not guarantee transactional rollback.

Unbounded work violates cancellation and budget expectations.

Partial generation has not passed completion checks.

Question 284

A prompt workflow has inconsistent per-stage formats. Which development or release-control change meets this requirement?

Accept any upstream free-text payload and rely on the final prompt to interpret it correctly.
Run mutating tools on unvalidated extraction results because the final summary will be checked.
Drop stage-level error statuses and always route to the final success node.
Validate the output schema at every stage, not solely final prose.

Explanation

Correct Answer

Validate typed outputs at each workflow stage, not just the final reply, so invalid intermediate data cannot trigger later actions.

Incorrect Answers

A later model cannot guarantee repaired semantics for bad intermediate fields.

A post-hoc text check cannot undo unsafe side effects.

Required failures need explicit handling rather than silent promotion.

Question 285

A GenAI deploy pipeline has no rollback for bad prompt changes. Which integration fix is appropriate for this failure?

Publish a new draft prompt directly to all users without retaining the previous tested version.
Evaluate only infrastructure health and ignore generated answer regressions.
Record versioned artifacts and automated regression-triggered rollback.
Overwrite all prompt acceptance fixtures after promotion so comparisons always pass.

Explanation

Correct Answer

Pin prompt/model artifacts to the release and keep tested rollback triggered by semantic, security or latency regression—not only deployment failures.

Incorrect Answers

There is no reliable rollback target if the new prompt misbehaves.

A healthy deployment can still misstate policy content.

Changing the baseline destroys evidence of regression.

Question 286

A public assistant is asked to provide prohibited self-harm instructions. Which enforcement decision best mitigates this failure?

Ask the model alone to decide whether dangerous instructions are appropriate without an enforced application policy.
Disable the guardrail for a user who frames the harmful question as research.
Apply policy-aligned Bedrock Guardrails content filtering with a safe blocked response.
Log prohibited prompts but still show detailed unsafe generated content to users.

Explanation

Correct Answer

Apply a tested content policy/guardrail to the actual input and/or output, returning the configured safe blocked message on prohibited harmful requests.

Incorrect Answers

Model self-moderation can fail on jailbreak or ambiguous requests.

Framing does not establish policy authorization to supply dangerous instructions.

Monitoring without intervention fails the stated blocking requirement.

Question 287

A financial chatbot responds to forbidden investment advice topics. What should be implemented before letting the model answer or act?

Use a topic named 'anything except finance' and assume the guardrail infers the intended exclusions.
Configure an appropriate denied-topic policy and test topic boundaries.
Filter only the word 'stock' instead of contextual investment recommendations.
Moderate user input only and allow the model to generate unsolicited investment recommendations.

Explanation

Correct Answer

Define investment advice as a positively worded denied topic and test both intended refusals and allowed neutral educational cases.

Incorrect Answers

Negative topic definitions are discouraged and ambiguous.

Keyword alone overblocks benign context and misses advice lacking that word.

A prohibited topic may appear in output even if the prompt itself was benign.

Question 288

A moderation rule uses negative-topic definitions to exempt some illegal content. Which safety control actually addresses the reported weakness?

Define the denied topic as 'all content except permissible content' in natural-language instructions.
Put 'do not mention X' as the topic definition and expect reliable classification.
Use the topic name as a regex list of prohibited entity strings.
Rewrite denied topics as clear positive subject definitions with tested exclusions outside the rule.

Explanation

Correct Answer

AWS recommends describing denied topics by affirmative subject meaning, not instructions, negated categories or exception-based definitions.

Incorrect Answers

A negative exclusion is not a clear positive topical definition.

The definition should describe the subject rather than command the filter.

Denied topics are contextual classifiers; word lists have separate features.

Question 289

A customer enters a national identification number. Which enforcement decision best mitigates this failure?

Rely on a general toxicity classifier to detect national identity numbers accurately.
Store the entire ID in logs first and redact only the generated reply.
Detect and mask or block PII using supported sensitive-information filters.
Replace all digits with zeros without validating whether they belong to the required ID field.

Explanation

Correct Answer

Use sensitive-information filtering with supported identifiers and policy-driven BLOCK or ANONYMIZE/MASK behavior for the national ID field.

Incorrect Answers

Toxicity categories are not a reliable PII detector.

Sensitive input remains exposed in storage.

Blind mutation can destroy needed non-sensitive data and lacks precise detection.

Question 290

An enterprise assistant receives an instruction to ignore its system policy and reveal another employee's account. Which protection combines an appropriate Bedrock prompt-attack filter with the essential independent access boundary?

Rely on a low model temperature as proof that the instruction-override attack will not succeed.
Configure supported Bedrock Guardrails prompt-attack detection and independently enforce caller authorization on protected data/tools.
Tell the attacker which sensitive records exist so the model can explain why it refuses.
Grant the model an administrator tool role but ask politely that it not use the role.

Explanation

Correct Answer

Guardrails can help detect prompt attacks, while the application or tool still must validate identity and record-level authorization regardless of model output.

Incorrect Answers

Sampling settings do not detect jailbreaks or enforce the caller's resource permissions.

Exposing the existence or details of restricted records can itself leak sensitive information.

A broad role remains broad even with benign system instructions; a prompt cannot substitute for least-privilege access controls.

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

Question 280

Explanation

Correct Answer

Incorrect Answers

Question 281

Explanation

Correct Answer

Incorrect Answers

Question 282

Explanation

Correct Answer

Incorrect Answers

Question 283

Explanation

Correct Answer

Incorrect Answers

Question 284

Explanation

Correct Answer

Incorrect Answers

Question 285

Explanation

Correct Answer

Incorrect Answers

Question 286

Explanation

Correct Answer

Incorrect Answers

Question 287

Explanation

Correct Answer

Incorrect Answers

Question 288

Explanation

Correct Answer

Incorrect Answers

Question 289

Explanation

Correct Answer

Incorrect Answers

Question 290

Explanation

Correct Answer

Incorrect Answers

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