Amazon AWS AIP-C01 Input and Output Safety Controls Practice Test

 

Topic 12 focuses on Input and Output Safety Controls for the AWS Certified Generative AI Developer – Professional certification and the AIP-C01 exam, using Amazon Bedrock and AWS generative AI services and architecture scenarios where relevant. For broader exam preparation, review the AWS Certified Generative AI Developer – Professional AIP-C01 Exam Dumps page. Each question includes a concise explanation of the correct answer and the technical reason the other choices are incorrect.

Question 1

The case management platform must block or transform harmful, disallowed, or sensitive user input before model processing. Which approach is the strongest fit?

1. Amazon Bedrock Guardrails input filtering
2. Lambda post-processing validation
3. custom moderation workflow with Step Functions and Lambda
4. Knowledge Bases grounding

Correct Answer: A

 

Correct Answer

Answer A is correct because Amazon Bedrock Guardrails input filtering is designed to block or transform harmful, disallowed, or sensitive user input before model processing. It applies configured content, denied-topic, word, or sensitive-information policies at the model interaction boundary.

Incorrect Answers

Answer B is incorrect because Lambda post-processing validation is primarily used to run custom checks on model output before returning or persisting it, which is a different requirement from the one being tested.

Answer C is incorrect because custom moderation workflow with Step Functions and Lambda is primarily used to apply organization-specific validation steps that are not covered by a single managed safety control, which is a different requirement from the one being tested.

Answer D is incorrect because Knowledge Bases grounding is primarily used to reduce unsupported answers by supplying retrieved source context to the model, which is a different requirement from the one being tested.

 

Question 2

The legal research platform must apply organization-specific validation steps that are not covered by a single managed safety control. Which approach is the strongest fit?

1. custom moderation workflow with Step Functions and Lambda
2. Amazon Bedrock Guardrails output filtering
3. confidence and semantic verification
4. API Gateway response filtering

Correct Answer: A

 

Correct Answer

Answer A is correct because custom moderation workflow with Step Functions and Lambda is designed to apply organization-specific validation steps that are not covered by a single managed safety control. It orchestrates deterministic checks, custom classifiers, and escalation logic before or after model invocation.

Incorrect Answers

Answer B is incorrect because Amazon Bedrock Guardrails output filtering is primarily used to prevent unsafe or policy-violating model responses from reaching the user, which is a different requirement from the one being tested.

Answer C is incorrect because confidence and semantic verification is primarily used to flag answers whose claims are weakly supported by retrieved context, which is a different requirement from the one being tested.

Answer D is incorrect because API Gateway response filtering is primarily used to apply a final API-boundary control before a generated response leaves the service, which is a different requirement from the one being tested.

 

Question 3

The research assistant application needs an architecture that can prevent unsafe or policy-violating model responses from reaching the user. Which choice best meets that need?

1. toxicity and harmful-content evaluation
2. Amazon Bedrock Guardrails output filtering
3. JSON Schema output validation
4. prompt-injection detection

Correct Answer: B

 

Correct Answer

Answer B is correct because Amazon Bedrock Guardrails output filtering is designed to prevent unsafe or policy-violating model responses from reaching the user. It evaluates generated output against configured safety and sensitive-information policies before delivery.

Incorrect Answers

Answer A is incorrect because toxicity and harmful-content evaluation is primarily used to measure whether model outputs violate content-safety thresholds, which is a different requirement from the one being tested.

Answer C is incorrect because JSON Schema output validation is primarily used to enforce machine-readable response structure before downstream automation acts on model output, which is a different requirement from the one being tested.

Answer D is incorrect because prompt-injection detection is primarily used to identify inputs that attempt to override system instructions or manipulate tool behavior, which is a different requirement from the one being tested.

 

Question 4

The AI product engineering team must measure whether model outputs violate content-safety thresholds. Which approach is the strongest fit?

1. Knowledge Bases grounding
2. toxicity and harmful-content evaluation
3. Amazon Comprehend preprocessing
4. automated jailbreak and adversarial testing

Correct Answer: B

 

Correct Answer

Answer B is correct because toxicity and harmful-content evaluation is designed to measure whether model outputs violate content-safety thresholds. It scores or classifies generated content so unsafe behavior can be blocked, compared, or tracked over time.

Incorrect Answers

Answer A is incorrect because Knowledge Bases grounding is primarily used to reduce unsupported answers by supplying retrieved source context to the model, which is a different requirement from the one being tested.

Answer C is incorrect because Amazon Comprehend preprocessing is primarily used to detect sensitive or risky text features before the prompt reaches the foundation model, which is a different requirement from the one being tested.

Answer D is incorrect because automated jailbreak and adversarial testing is primarily used to continuously test safety controls against intentionally manipulative prompts, which is a different requirement from the one being tested.

 

Question 5

The supply-chain analytics group has a design goal to reduce unsupported answers by supplying retrieved source context to the model. What should the team choose?

1. confidence and semantic verification
2. Lambda post-processing validation
3. Amazon Bedrock Guardrails input filtering
4. Knowledge Bases grounding

Correct Answer: D

 

Correct Answer

Answer D is correct because Knowledge Bases grounding is designed to reduce unsupported answers by supplying retrieved source context to the model. It anchors generation in enterprise information retrieved from an approved knowledge corpus instead of relying only on parametric memory.

Incorrect Answers

Answer A is incorrect because confidence and semantic verification is primarily used to flag answers whose claims are weakly supported by retrieved context, which is a different requirement from the one being tested.

Answer B is incorrect because Lambda post-processing validation is primarily used to run custom checks on model output before returning or persisting it, which is a different requirement from the one being tested.

Answer C is incorrect because Amazon Bedrock Guardrails input filtering is primarily used to block or transform harmful, disallowed, or sensitive user input before model processing, which is a different requirement from the one being tested.

 

Question 6

The enterprise developer platform must flag answers whose claims are weakly supported by retrieved context. Which approach is the strongest fit?

1. confidence and semantic verification
2. custom moderation workflow with Step Functions and Lambda
3. API Gateway response filtering
4. JSON Schema output validation

Correct Answer: A

 

Correct Answer

Answer A is correct because confidence and semantic verification is designed to flag answers whose claims are weakly supported by retrieved context. It compares response content with evidence or confidence signals before accepting the output.

Incorrect Answers

Answer B is incorrect because custom moderation workflow with Step Functions and Lambda is primarily used to apply organization-specific validation steps that are not covered by a single managed safety control, which is a different requirement from the one being tested.

Answer C is incorrect because API Gateway response filtering is primarily used to apply a final API-boundary control before a generated response leaves the service, which is a different requirement from the one being tested.

Answer D is incorrect because JSON Schema output validation is primarily used to enforce machine-readable response structure before downstream automation acts on model output, which is a different requirement from the one being tested.

 

Question 7

The customer identity platform must enforce machine-readable response structure before downstream automation acts on model output. Which approach is the strongest fit?

1. JSON Schema output validation
2. Amazon Bedrock Guardrails output filtering
3. Amazon Comprehend preprocessing
4. prompt-injection detection

Correct Answer: A

 

Correct Answer

Answer A is correct because JSON Schema output validation is designed to enforce machine-readable response structure before downstream automation acts on model output. It checks that generated fields and types conform to an expected schema and rejects malformed results.

Incorrect Answers

Answer B is incorrect because Amazon Bedrock Guardrails output filtering is primarily used to prevent unsafe or policy-violating model responses from reaching the user, which is a different requirement from the one being tested.

Answer C is incorrect because Amazon Comprehend preprocessing is primarily used to detect sensitive or risky text features before the prompt reaches the foundation model, which is a different requirement from the one being tested.

Answer D is incorrect because prompt-injection detection is primarily used to identify inputs that attempt to override system instructions or manipulate tool behavior, which is a different requirement from the one being tested.

 

Question 8

The business intelligence application needs an architecture that can detect sensitive or risky text features before the prompt reaches the foundation model. Which choice best meets that need?

1. Lambda post-processing validation
2. Amazon Comprehend preprocessing
3. automated jailbreak and adversarial testing
4. toxicity and harmful-content evaluation

Correct Answer: B

 

Correct Answer

Answer B is correct because Amazon Comprehend preprocessing is designed to detect sensitive or risky text features before the prompt reaches the foundation model. It performs managed NLP analysis that can feed a deterministic input-filtering decision.

Incorrect Answers

Answer A is incorrect because Lambda post-processing validation is primarily used to run custom checks on model output before returning or persisting it, which is a different requirement from the one being tested.

Answer C is incorrect because automated jailbreak and adversarial testing is primarily used to continuously test safety controls against intentionally manipulative prompts, which is a different requirement from the one being tested.

Answer D is incorrect because toxicity and harmful-content evaluation is primarily used to measure whether model outputs violate content-safety thresholds, which is a different requirement from the one being tested.

 

Question 9

The healthcare document platform has a design goal to run custom checks on model output before returning or persisting it. What should the team choose?

1. Amazon Bedrock Guardrails input filtering
2. Knowledge Bases grounding
3. Lambda post-processing validation
4. API Gateway response filtering

Correct Answer: C

 

Correct Answer

Answer C is correct because Lambda post-processing validation is designed to run custom checks on model output before returning or persisting it. It executes organization-specific rules, transformations, or policy validation after inference.

Incorrect Answers

Answer A is incorrect because Amazon Bedrock Guardrails input filtering is primarily used to block or transform harmful, disallowed, or sensitive user input before model processing, which is a different requirement from the one being tested.

Answer B is incorrect because Knowledge Bases grounding is primarily used to reduce unsupported answers by supplying retrieved source context to the model, which is a different requirement from the one being tested.

Answer D is incorrect because API Gateway response filtering is primarily used to apply a final API-boundary control before a generated response leaves the service, which is a different requirement from the one being tested.

 

Question 10

The e-commerce search team has a design goal to apply a final API-boundary control before a generated response leaves the service. What should the team choose?

1. confidence and semantic verification
2. custom moderation workflow with Step Functions and Lambda
3. API Gateway response filtering
4. prompt-injection detection

Correct Answer: C

 

Correct Answer

Answer C is correct because API Gateway response filtering is designed to apply a final API-boundary control before a generated response leaves the service. It uses gateway or integration logic to enforce response transformations and policy checks at the external interface.

Incorrect Answers

Answer A is incorrect because confidence and semantic verification is primarily used to flag answers whose claims are weakly supported by retrieved context, which is a different requirement from the one being tested.

Answer B is incorrect because custom moderation workflow with Step Functions and Lambda is primarily used to apply organization-specific validation steps that are not covered by a single managed safety control, which is a different requirement from the one being tested.

Answer D is incorrect because prompt-injection detection is primarily used to identify inputs that attempt to override system instructions or manipulate tool behavior, which is a different requirement from the one being tested.

 

Question 11

The risk analytics team has a design goal to identify inputs that attempt to override system instructions or manipulate tool behavior. What should the team choose?

1. prompt-injection detection
2. Amazon Bedrock Guardrails output filtering
3. JSON Schema output validation
4. automated jailbreak and adversarial testing

Correct Answer: A

 

Correct Answer

Answer A is correct because prompt-injection detection is designed to identify inputs that attempt to override system instructions or manipulate tool behavior. It looks for adversarial instruction patterns and suspicious context transitions before trusting the request.

Incorrect Answers

Answer B is incorrect because Amazon Bedrock Guardrails output filtering is primarily used to prevent unsafe or policy-violating model responses from reaching the user, which is a different requirement from the one being tested.

Answer C is incorrect because JSON Schema output validation is primarily used to enforce machine-readable response structure before downstream automation acts on model output, which is a different requirement from the one being tested.

Answer D is incorrect because automated jailbreak and adversarial testing is primarily used to continuously test safety controls against intentionally manipulative prompts, which is a different requirement from the one being tested.

 

Question 12

The fraud detection engineering team is prioritizing a requirement to continuously test safety controls against intentionally manipulative prompts. Which implementation is most appropriate?

1. Amazon Bedrock Guardrails input filtering
2. Amazon Comprehend preprocessing
3. automated jailbreak and adversarial testing
4. toxicity and harmful-content evaluation

Correct Answer: C

 

Correct Answer

Answer C is correct because automated jailbreak and adversarial testing is designed to continuously test safety controls against intentionally manipulative prompts. It runs representative attack suites so control regressions are detected before or during production operation.

Incorrect Answers

Answer A is incorrect because Amazon Bedrock Guardrails input filtering is primarily used to block or transform harmful, disallowed, or sensitive user input before model processing, which is a different requirement from the one being tested.

Answer B is incorrect because Amazon Comprehend preprocessing is primarily used to detect sensitive or risky text features before the prompt reaches the foundation model, which is a different requirement from the one being tested.

Answer D is incorrect because toxicity and harmful-content evaluation is primarily used to measure whether model outputs violate content-safety thresholds, which is a different requirement from the one being tested.

 

Question 13

The field-service application documents this GenAI behavior: applies configured content, denied-topic, word, or sensitive-information policies at the model interaction boundary. Which capability matches it?

1. Knowledge Bases grounding
2. Amazon Bedrock Guardrails input filtering
3. custom moderation workflow with Step Functions and Lambda
4. Lambda post-processing validation

Correct Answer: B

 

Correct Answer

Answer B is correct because the description directly matches Amazon Bedrock Guardrails input filtering. Amazon Bedrock Guardrails input filtering applies configured content, denied-topic, word, or sensitive-information policies at the model interaction boundary.

Incorrect Answers

Answer A is incorrect because Knowledge Bases grounding is primarily used to reduce unsupported answers by supplying retrieved source context to the model, which is a different requirement from the one being tested.

Answer C is incorrect because custom moderation workflow with Step Functions and Lambda is primarily used to apply organization-specific validation steps that are not covered by a single managed safety control, which is a different requirement from the one being tested.

Answer D is incorrect because Lambda post-processing validation is primarily used to run custom checks on model output before returning or persisting it, which is a different requirement from the one being tested.

 

Question 14

The model governance committee describes a component that orchestrates deterministic checks, custom classifiers, and escalation logic before or after model invocation. Which capability is being described?

1. confidence and semantic verification
2. custom moderation workflow with Step Functions and Lambda
3. Amazon Bedrock Guardrails output filtering
4. API Gateway response filtering

Correct Answer: B

 

Correct Answer

Answer B is correct because the description directly matches custom moderation workflow with Step Functions and Lambda. custom moderation workflow with Step Functions and Lambda orchestrates deterministic checks, custom classifiers, and escalation logic before or after model invocation.

Incorrect Answers

Answer A is incorrect because confidence and semantic verification is primarily used to flag answers whose claims are weakly supported by retrieved context, which is a different requirement from the one being tested.

Answer C is incorrect because Amazon Bedrock Guardrails output filtering is primarily used to prevent unsafe or policy-violating model responses from reaching the user, which is a different requirement from the one being tested.

Answer D is incorrect because API Gateway response filtering is primarily used to apply a final API-boundary control before a generated response leaves the service, which is a different requirement from the one being tested.

 

Question 15

The AI governance program describes a component that evaluates generated output against configured safety and sensitive-information policies before delivery. Which capability is being described?

1. prompt-injection detection
2. toxicity and harmful-content evaluation
3. JSON Schema output validation
4. Amazon Bedrock Guardrails output filtering

Correct Answer: D

 

Correct Answer

Answer D is correct because the description directly matches Amazon Bedrock Guardrails output filtering. Amazon Bedrock Guardrails output filtering evaluates generated output against configured safety and sensitive-information policies before delivery.

Incorrect Answers

Answer A is incorrect because prompt-injection detection is primarily used to identify inputs that attempt to override system instructions or manipulate tool behavior, which is a different requirement from the one being tested.

Answer B is incorrect because toxicity and harmful-content evaluation is primarily used to measure whether model outputs violate content-safety thresholds, which is a different requirement from the one being tested.

Answer C is incorrect because JSON Schema output validation is primarily used to enforce machine-readable response structure before downstream automation acts on model output, which is a different requirement from the one being tested.

 

Question 16

Within the sales operations automation team’s architecture, which capability matches this technical description: scores or classifies generated content so unsafe behavior can be blocked, compared, or tracked over time?

1. toxicity and harmful-content evaluation
2. automated jailbreak and adversarial testing
3. Amazon Comprehend preprocessing
4. Knowledge Bases grounding

Correct Answer: A

 

Correct Answer

Answer A is correct because the description directly matches toxicity and harmful-content evaluation. toxicity and harmful-content evaluation scores or classifies generated content so unsafe behavior can be blocked, compared, or tracked over time.

Incorrect Answers

Answer B is incorrect because automated jailbreak and adversarial testing is primarily used to continuously test safety controls against intentionally manipulative prompts, which is a different requirement from the one being tested.

Answer C is incorrect because Amazon Comprehend preprocessing is primarily used to detect sensitive or risky text features before the prompt reaches the foundation model, which is a different requirement from the one being tested.

Answer D is incorrect because Knowledge Bases grounding is primarily used to reduce unsupported answers by supplying retrieved source context to the model, which is a different requirement from the one being tested.

 

Question 17

Within the knowledge retrieval engineering team’s architecture, which capability matches this technical description: anchors generation in enterprise information retrieved from an approved knowledge corpus instead of relying only on parametric memory?

1. confidence and semantic verification
2. Lambda post-processing validation
3. Knowledge Bases grounding
4. Amazon Bedrock Guardrails input filtering

Correct Answer: C

 

Correct Answer

Answer C is correct because the description directly matches Knowledge Bases grounding. Knowledge Bases grounding anchors generation in enterprise information retrieved from an approved knowledge corpus instead of relying only on parametric memory.

Incorrect Answers

Answer A is incorrect because confidence and semantic verification is primarily used to flag answers whose claims are weakly supported by retrieved context, which is a different requirement from the one being tested.

Answer B is incorrect because Lambda post-processing validation is primarily used to run custom checks on model output before returning or persisting it, which is a different requirement from the one being tested.

Answer D is incorrect because Amazon Bedrock Guardrails input filtering is primarily used to block or transform harmful, disallowed, or sensitive user input before model processing, which is a different requirement from the one being tested.

 

Question 18

Within the enterprise content repository team’s architecture, which capability matches this technical description: compares response content with evidence or confidence signals before accepting the output?

1. API Gateway response filtering
2. JSON Schema output validation
3. custom moderation workflow with Step Functions and Lambda
4. confidence and semantic verification

Correct Answer: D

 

Correct Answer

Answer D is correct because the description directly matches confidence and semantic verification. confidence and semantic verification compares response content with evidence or confidence signals before accepting the output.

Incorrect Answers

Answer A is incorrect because API Gateway response filtering is primarily used to apply a final API-boundary control before a generated response leaves the service, which is a different requirement from the one being tested.

Answer B is incorrect because JSON Schema output validation is primarily used to enforce machine-readable response structure before downstream automation acts on model output, which is a different requirement from the one being tested.

Answer C is incorrect because custom moderation workflow with Step Functions and Lambda is primarily used to apply organization-specific validation steps that are not covered by a single managed safety control, which is a different requirement from the one being tested.

 

Question 19

The manufacturing analytics team needs to identify a capability with this behavior: checks that generated fields and types conform to an expected schema and rejects malformed results. What is the best match?

1. prompt-injection detection
2. Amazon Bedrock Guardrails output filtering
3. JSON Schema output validation
4. Amazon Comprehend preprocessing

Correct Answer: C

 

Correct Answer

Answer C is correct because the description directly matches JSON Schema output validation. JSON Schema output validation checks that generated fields and types conform to an expected schema and rejects malformed results.

Incorrect Answers

Answer A is incorrect because prompt-injection detection is primarily used to identify inputs that attempt to override system instructions or manipulate tool behavior, which is a different requirement from the one being tested.

Answer B is incorrect because Amazon Bedrock Guardrails output filtering is primarily used to prevent unsafe or policy-violating model responses from reaching the user, which is a different requirement from the one being tested.

Answer D is incorrect because Amazon Comprehend preprocessing is primarily used to detect sensitive or risky text features before the prompt reaches the foundation model, which is a different requirement from the one being tested.

 

Question 20

The developer productivity team needs to identify a capability with this behavior: performs managed NLP analysis that can feed a deterministic input-filtering decision. What is the best match?

1. Lambda post-processing validation
2. automated jailbreak and adversarial testing
3. Amazon Comprehend preprocessing
4. toxicity and harmful-content evaluation

Correct Answer: C

 

Correct Answer

Answer C is correct because the description directly matches Amazon Comprehend preprocessing. Amazon Comprehend preprocessing performs managed NLP analysis that can feed a deterministic input-filtering decision.

Incorrect Answers

Answer A is incorrect because Lambda post-processing validation is primarily used to run custom checks on model output before returning or persisting it, which is a different requirement from the one being tested.

Answer B is incorrect because automated jailbreak and adversarial testing is primarily used to continuously test safety controls against intentionally manipulative prompts, which is a different requirement from the one being tested.

Answer D is incorrect because toxicity and harmful-content evaluation is primarily used to measure whether model outputs violate content-safety thresholds, which is a different requirement from the one being tested.

 

Question 21

Within the business process automation team’s architecture, which capability matches this technical description: executes organization-specific rules, transformations, or policy validation after inference?

1. Lambda post-processing validation
2. Knowledge Bases grounding
3. API Gateway response filtering
4. Amazon Bedrock Guardrails input filtering

Correct Answer: A

 

Correct Answer

Answer A is correct because the description directly matches Lambda post-processing validation. Lambda post-processing validation executes organization-specific rules, transformations, or policy validation after inference.

Incorrect Answers

Answer B is incorrect because Knowledge Bases grounding is primarily used to reduce unsupported answers by supplying retrieved source context to the model, which is a different requirement from the one being tested.

Answer C is incorrect because API Gateway response filtering is primarily used to apply a final API-boundary control before a generated response leaves the service, which is a different requirement from the one being tested.

Answer D is incorrect because Amazon Bedrock Guardrails input filtering is primarily used to block or transform harmful, disallowed, or sensitive user input before model processing, which is a different requirement from the one being tested.

 

Question 22

The regulatory reporting platform documents this GenAI behavior: uses gateway or integration logic to enforce response transformations and policy checks at the external interface. Which capability matches it?

1. confidence and semantic verification
2. API Gateway response filtering
3. custom moderation workflow with Step Functions and Lambda
4. prompt-injection detection

Correct Answer: B

 

Correct Answer

Answer B is correct because the description directly matches API Gateway response filtering. API Gateway response filtering uses gateway or integration logic to enforce response transformations and policy checks at the external interface.

Incorrect Answers

Answer A is incorrect because confidence and semantic verification is primarily used to flag answers whose claims are weakly supported by retrieved context, which is a different requirement from the one being tested.

Answer C is incorrect because custom moderation workflow with Step Functions and Lambda is primarily used to apply organization-specific validation steps that are not covered by a single managed safety control, which is a different requirement from the one being tested.

Answer D is incorrect because prompt-injection detection is primarily used to identify inputs that attempt to override system instructions or manipulate tool behavior, which is a different requirement from the one being tested.

 

Question 23

The privacy engineering function documents this GenAI behavior: looks for adversarial instruction patterns and suspicious context transitions before trusting the request. Which capability matches it?

1. Amazon Bedrock Guardrails output filtering
2. prompt-injection detection
3. automated jailbreak and adversarial testing
4. JSON Schema output validation

Correct Answer: B

 

Correct Answer

Answer B is correct because the description directly matches prompt-injection detection. prompt-injection detection looks for adversarial instruction patterns and suspicious context transitions before trusting the request.

Incorrect Answers

Answer A is incorrect because Amazon Bedrock Guardrails output filtering is primarily used to prevent unsafe or policy-violating model responses from reaching the user, which is a different requirement from the one being tested.

Answer C is incorrect because automated jailbreak and adversarial testing is primarily used to continuously test safety controls against intentionally manipulative prompts, which is a different requirement from the one being tested.

Answer D is incorrect because JSON Schema output validation is primarily used to enforce machine-readable response structure before downstream automation acts on model output, which is a different requirement from the one being tested.

 

Question 24

Within the logistics optimization team’s architecture, which capability matches this technical description: runs representative attack suites so control regressions are detected before or during production operation?

1. toxicity and harmful-content evaluation
2. Amazon Bedrock Guardrails input filtering
3. Amazon Comprehend preprocessing
4. automated jailbreak and adversarial testing

Correct Answer: D

 

Correct Answer

Answer D is correct because the description directly matches automated jailbreak and adversarial testing. automated jailbreak and adversarial testing runs representative attack suites so control regressions are detected before or during production operation.

Incorrect Answers

Answer A is incorrect because toxicity and harmful-content evaluation is primarily used to measure whether model outputs violate content-safety thresholds, which is a different requirement from the one being tested.

Answer B is incorrect because Amazon Bedrock Guardrails input filtering is primarily used to block or transform harmful, disallowed, or sensitive user input before model processing, which is a different requirement from the one being tested.

Answer C is incorrect because Amazon Comprehend preprocessing is primarily used to detect sensitive or risky text features before the prompt reaches the foundation model, which is a different requirement from the one being tested.

 

Question 25

Why would the customer analytics platform introduce Amazon Bedrock Guardrails input filtering into the GenAI architecture?

1. Run custom checks on model output before returning or persisting it
2. Apply organization-specific validation steps that are not covered by a single managed safety control
3. Block or transform harmful, disallowed, or sensitive user input before model processing
4. Reduce unsupported answers by supplying retrieved source context to the model

Correct Answer: C

 

Correct Answer

Answer C is correct because Amazon Bedrock Guardrails input filtering is specifically used to block or transform harmful, disallowed, or sensitive user input before model processing. It applies configured content, denied-topic, word, or sensitive-information policies at the model interaction boundary.

Incorrect Answers

Answer A is incorrect because that requirement aligns with Lambda post-processing validation, not Amazon Bedrock Guardrails input filtering.

Answer B is incorrect because that requirement aligns with custom moderation workflow with Step Functions and Lambda, not Amazon Bedrock Guardrails input filtering.

Answer D is incorrect because that requirement aligns with Knowledge Bases grounding, not Amazon Bedrock Guardrails input filtering.

 

Question 26

The enterprise finance automation team is considering custom moderation workflow with Step Functions and Lambda. What problem is this choice primarily meant to solve?

1. Flag answers whose claims are weakly supported by retrieved context
2. Apply a final API-boundary control before a generated response leaves the service
3. Prevent unsafe or policy-violating model responses from reaching the user
4. Apply organization-specific validation steps that are not covered by a single managed safety control

Correct Answer: D

 

Correct Answer

Answer D is correct because custom moderation workflow with Step Functions and Lambda is specifically used to apply organization-specific validation steps that are not covered by a single managed safety control. It orchestrates deterministic checks, custom classifiers, and escalation logic before or after model invocation.

Incorrect Answers

Answer A is incorrect because that requirement aligns with confidence and semantic verification, not custom moderation workflow with Step Functions and Lambda.

Answer B is incorrect because that requirement aligns with API Gateway response filtering, not custom moderation workflow with Step Functions and Lambda.

Answer C is incorrect because that requirement aligns with Amazon Bedrock Guardrails output filtering, not custom moderation workflow with Step Functions and Lambda.

 

Question 27

The risk and controls team is considering Amazon Bedrock Guardrails output filtering. What problem is this choice primarily meant to solve?

1. Identify inputs that attempt to override system instructions or manipulate tool behavior
2. Prevent unsafe or policy-violating model responses from reaching the user
3. Enforce machine-readable response structure before downstream automation acts on model output
4. Measure whether model outputs violate content-safety thresholds

Correct Answer: B

 

Correct Answer

Answer B is correct because Amazon Bedrock Guardrails output filtering is specifically used to prevent unsafe or policy-violating model responses from reaching the user. It evaluates generated output against configured safety and sensitive-information policies before delivery.

Incorrect Answers

Answer A is incorrect because that requirement aligns with prompt-injection detection, not Amazon Bedrock Guardrails output filtering.

Answer C is incorrect because that requirement aligns with JSON Schema output validation, not Amazon Bedrock Guardrails output filtering.

Answer D is incorrect because that requirement aligns with toxicity and harmful-content evaluation, not Amazon Bedrock Guardrails output filtering.

 

Question 28

The enterprise search architecture group is considering toxicity and harmful-content evaluation. What problem is this choice primarily meant to solve?

1. Measure whether model outputs violate content-safety thresholds
2. Detect sensitive or risky text features before the prompt reaches the foundation model
3. Reduce unsupported answers by supplying retrieved source context to the model
4. Continuously test safety controls against intentionally manipulative prompts

Correct Answer: A

 

Correct Answer

Answer A is correct because toxicity and harmful-content evaluation is specifically used to measure whether model outputs violate content-safety thresholds. It scores or classifies generated content so unsafe behavior can be blocked, compared, or tracked over time.

Incorrect Answers

Answer B is incorrect because that requirement aligns with Amazon Comprehend preprocessing, not toxicity and harmful-content evaluation.

Answer C is incorrect because that requirement aligns with Knowledge Bases grounding, not toxicity and harmful-content evaluation.

Answer D is incorrect because that requirement aligns with automated jailbreak and adversarial testing, not toxicity and harmful-content evaluation.

 

Question 29

Why would the insurance automation team introduce Knowledge Bases grounding into the GenAI architecture?

1. Reduce unsupported answers by supplying retrieved source context to the model
2. Block or transform harmful, disallowed, or sensitive user input before model processing
3. Flag answers whose claims are weakly supported by retrieved context
4. Run custom checks on model output before returning or persisting it

Correct Answer: A

 

Correct Answer

Answer A is correct because Knowledge Bases grounding is specifically used to reduce unsupported answers by supplying retrieved source context to the model. It anchors generation in enterprise information retrieved from an approved knowledge corpus instead of relying only on parametric memory.

Incorrect Answers

Answer B is incorrect because that requirement aligns with Amazon Bedrock Guardrails input filtering, not Knowledge Bases grounding.

Answer C is incorrect because that requirement aligns with confidence and semantic verification, not Knowledge Bases grounding.

Answer D is incorrect because that requirement aligns with Lambda post-processing validation, not Knowledge Bases grounding.

 

Question 30

The procurement automation group plans to adopt confidence and semantic verification. Which outcome should drive that decision?

1. Apply a final API-boundary control before a generated response leaves the service
2. Enforce machine-readable response structure before downstream automation acts on model output
3. Apply organization-specific validation steps that are not covered by a single managed safety control
4. Flag answers whose claims are weakly supported by retrieved context

Correct Answer: D

 

Correct Answer

Answer D is correct because confidence and semantic verification is specifically used to flag answers whose claims are weakly supported by retrieved context. It compares response content with evidence or confidence signals before accepting the output.

Incorrect Answers

Answer A is incorrect because that requirement aligns with API Gateway response filtering, not confidence and semantic verification.

Answer B is incorrect because that requirement aligns with JSON Schema output validation, not confidence and semantic verification.

Answer C is incorrect because that requirement aligns with custom moderation workflow with Step Functions and Lambda, not confidence and semantic verification.

 

Question 31

The observability engineering group plans to adopt JSON Schema output validation. Which outcome should drive that decision?

1. Prevent unsafe or policy-violating model responses from reaching the user
2. Identify inputs that attempt to override system instructions or manipulate tool behavior
3. Enforce machine-readable response structure before downstream automation acts on model output
4. Detect sensitive or risky text features before the prompt reaches the foundation model

Correct Answer: C

 

Correct Answer

Answer C is correct because JSON Schema output validation is specifically used to enforce machine-readable response structure before downstream automation acts on model output. It checks that generated fields and types conform to an expected schema and rejects malformed results.

Incorrect Answers

Answer A is incorrect because that requirement aligns with Amazon Bedrock Guardrails output filtering, not JSON Schema output validation.

Answer B is incorrect because that requirement aligns with prompt-injection detection, not JSON Schema output validation.

Answer D is incorrect because that requirement aligns with Amazon Comprehend preprocessing, not JSON Schema output validation.

 

Question 32

Within the managed services operations team’s design, the team highlights Amazon Comprehend preprocessing. Which need does that component address?

1. Continuously test safety controls against intentionally manipulative prompts
2. Detect sensitive or risky text features before the prompt reaches the foundation model
3. Measure whether model outputs violate content-safety thresholds
4. Run custom checks on model output before returning or persisting it

Correct Answer: B

 

Correct Answer

Answer B is correct because Amazon Comprehend preprocessing is specifically used to detect sensitive or risky text features before the prompt reaches the foundation model. It performs managed NLP analysis that can feed a deterministic input-filtering decision.

Incorrect Answers

Answer A is incorrect because that requirement aligns with automated jailbreak and adversarial testing, not Amazon Comprehend preprocessing.

Answer C is incorrect because that requirement aligns with toxicity and harmful-content evaluation, not Amazon Comprehend preprocessing.

Answer D is incorrect because that requirement aligns with Lambda post-processing validation, not Amazon Comprehend preprocessing.

 

Question 33

The service management platform has proposed Lambda post-processing validation for its design. Which requirement best justifies it?

1. Apply a final API-boundary control before a generated response leaves the service
2. Reduce unsupported answers by supplying retrieved source context to the model
3. Block or transform harmful, disallowed, or sensitive user input before model processing
4. Run custom checks on model output before returning or persisting it

Correct Answer: D

 

Correct Answer

Answer D is correct because Lambda post-processing validation is specifically used to run custom checks on model output before returning or persisting it. It executes organization-specific rules, transformations, or policy validation after inference.

Incorrect Answers

Answer A is incorrect because that requirement aligns with API Gateway response filtering, not Lambda post-processing validation.

Answer B is incorrect because that requirement aligns with Knowledge Bases grounding, not Lambda post-processing validation.

Answer C is incorrect because that requirement aligns with Amazon Bedrock Guardrails input filtering, not Lambda post-processing validation.

 

Question 34

The digital experience platform is considering API Gateway response filtering. What problem is this choice primarily meant to solve?

1. Apply organization-specific validation steps that are not covered by a single managed safety control
2. Identify inputs that attempt to override system instructions or manipulate tool behavior
3. Apply a final API-boundary control before a generated response leaves the service
4. Flag answers whose claims are weakly supported by retrieved context

Correct Answer: C

 

Correct Answer

Answer C is correct because API Gateway response filtering is specifically used to apply a final API-boundary control before a generated response leaves the service. It uses gateway or integration logic to enforce response transformations and policy checks at the external interface.

Incorrect Answers

Answer A is incorrect because that requirement aligns with custom moderation workflow with Step Functions and Lambda, not API Gateway response filtering.

Answer B is incorrect because that requirement aligns with prompt-injection detection, not API Gateway response filtering.

Answer D is incorrect because that requirement aligns with confidence and semantic verification, not API Gateway response filtering.

 

Question 35

The cloud governance office is considering prompt-injection detection. What problem is this choice primarily meant to solve?

1. Enforce machine-readable response structure before downstream automation acts on model output
2. Prevent unsafe or policy-violating model responses from reaching the user
3. Continuously test safety controls against intentionally manipulative prompts
4. Identify inputs that attempt to override system instructions or manipulate tool behavior

Correct Answer: D

 

Correct Answer

Answer D is correct because prompt-injection detection is specifically used to identify inputs that attempt to override system instructions or manipulate tool behavior. It looks for adversarial instruction patterns and suspicious context transitions before trusting the request.

Incorrect Answers

Answer A is incorrect because that requirement aligns with JSON Schema output validation, not prompt-injection detection.

Answer B is incorrect because that requirement aligns with Amazon Bedrock Guardrails output filtering, not prompt-injection detection.

Answer C is incorrect because that requirement aligns with automated jailbreak and adversarial testing, not prompt-injection detection.