IAPP AIGP Dumps
IAPP AIGP Practice Test 4
Question 30
Choose the intervention and validation that fit this business problem: A legal intake team observes slightly different summaries when the same document is submitted twice.
Explanation
Correct Answer
For the action and verification, set repeatability tolerances and sample-based tests for a probabilistic AI service instead of requiring identical wording on every run. Run multiple controlled prompts and record factual accuracy, policy compliance and acceptable variance. Together, they implement and test the intended safeguard. Some generative systems vary their responses across runs; safe performance should be measured against stable quality and policy criteria
Incorrect Answers
This pair supports another governance intervention: Distinguish learned statistical prediction from a fixed rules engine before assigning AI-specific controls. For the actual case, the required action is: Set repeatability tolerances and sample-based tests for a probabilistic AI service instead of requiring identical wording on every run. The supporting evidence is: Run multiple controlled prompts and record factual accuracy, policy compliance and acceptable variance.
This pair supports another governance intervention: Classify a text-generating model separately from a numerical risk predictor when preparing evaluations. For the actual case, the required action is: Set repeatability tolerances and sample-based tests for a probabilistic AI service instead of requiring identical wording on every run. The supporting evidence is: Run multiple controlled prompts and record factual accuracy, policy compliance and acceptable variance.
This pair supports another governance intervention: Define triggers, reporting channels, containment authority and post-incident learning for AI harms. For the actual case, the required action is: Set repeatability tolerances and sample-based tests for a probabilistic AI service instead of requiring identical wording on every run. The supporting evidence is: Run multiple controlled prompts and record factual accuracy, policy compliance and acceptable variance.
Question 31
Consider the following proposed AI use: A housing nonprofit sees high average screening accuracy but unusually high rejection errors for one applicant group. Which response best addresses the governance gap?
Explanation
Correct Answer
Compare relevant error outcomes across affected groups and investigate material disparities before deployment. Aggregate accuracy can hide systematic harms to subpopulations even without explicit use of a protected attribute. The practical acceptance evidence is: review group-level false-positive and false-negative rates with population context and document mitigation decisions.
Incorrect Answers
This would help address a separate governance task: Distinguish learned statistical prediction from a fixed rules engine before assigning AI-specific controls. For this case, however, aggregate accuracy can hide systematic harms to subpopulations even without explicit use of a protected attribute. The more direct step is: Compare relevant error outcomes across affected groups and investigate material disparities before deployment.
This would help address a separate governance task: Assess confidentiality, data minimization and adversarial misuse together in the AI risk register. For this case, however, aggregate accuracy can hide systematic harms to subpopulations even without explicit use of a protected attribute. The more direct step is: Compare relevant error outcomes across affected groups and investigate material disparities before deployment.
This would help address a separate governance task: Train developers, approvers, operators and frontline users on the distinct AI risks and decisions they control. For this case, however, aggregate accuracy can hide systematic harms to subpopulations even without explicit use of a protected attribute. The more direct step is: Compare relevant error outcomes across affected groups and investigate material disparities before deployment.
Question 32
A housing nonprofit sees high average screening accuracy but unusually high rejection errors for one applicant group. Which explanation best connects the risk to the proposed governance action?
Explanation
Correct Answer
Aggregate accuracy can hide systematic harms to subpopulations even without explicit use of a protected attribute. That is why the case calls for this action: compare relevant error outcomes across affected groups and investigate material disparities before deployment.
Incorrect Answers
That explains why a team might classify a text-generating model separately from a numerical risk predictor when preparing evaluations, but it does not justify the decision required by the case. Aggregate accuracy can hide systematic harms to subpopulations even without explicit use of a protected attribute.
That explains why a team might distinguish learned statistical prediction from a fixed rules engine before assigning AI-specific controls, but it does not justify the decision required by the case. Aggregate accuracy can hide systematic harms to subpopulations even without explicit use of a protected attribute.
That explains why a team might attach explicit review and approval conditions to design, test, release and monitoring milestones, but it does not justify the decision required by the case. Aggregate accuracy can hide systematic harms to subpopulations even without explicit use of a protected attribute.
Question 33
A housing nonprofit sees high average screening accuracy but unusually high rejection errors for one applicant group. After taking the recommended governance action, which evidence most directly verifies that the risk has been addressed?
Explanation
Correct Answer
Review group-level false-positive and false-negative rates with population context and document mitigation decisions. This verifies the necessary safeguard rather than merely showing that an AI project was completed. Aggregate accuracy can hide systematic harms to subpopulations even without explicit use of a protected attribute.
Incorrect Answers
Test access controls, prompt abuse, output leakage and minimization against approved data-use boundaries would test a different outcome. The directly relevant evidence is to review group-level false-positive and false-negative rates with population context and document mitigation decisions.
Review samples of generated text for supported claims and separately test predictor error on held-out cases would test a different outcome. The directly relevant evidence is to review group-level false-positive and false-negative rates with population context and document mitigation decisions.
Use role-based exercises and verify that participants can identify triggers for escalation and human review would test a different outcome. The directly relevant evidence is to review group-level false-positive and false-negative rates with population context and document mitigation decisions.
Question 34
For this case—A housing nonprofit sees high average screening accuracy but unusually high rejection errors for one applicant group—staff choose the shortcut: Approve a screening model solely because its overall accuracy exceeds the headline target. Which consequence best explains why this is inadequate?
Explanation
Correct Answer
A well-performing average can conceal severe errors concentrated in one affected group. The shortcut overlooks the central safeguard: compare relevant error outcomes across affected groups and investigate material disparities before deployment. Aggregate accuracy can hide systematic harms to subpopulations even without explicit use of a protected attribute.
Incorrect Answers
This consequence follows more directly from another shortcut: Treat one satisfactory model response as proof that all future responses will be equally safe. The shortcut actually described would instead create this risk: A well-performing average can conceal severe errors concentrated in one affected group.
This consequence follows more directly from another shortcut: Label every software rule as machine learning and apply a single undifferentiated test checklist. The shortcut actually described would instead create this risk: A well-performing average can conceal severe errors concentrated in one affected group.
This consequence follows more directly from another shortcut: Allow teams to build a model first and identify its intended users only at launch. The shortcut actually described would instead create this risk: A well-performing average can conceal severe errors concentrated in one affected group.
Question 35
Choose the intervention and validation that fit this business problem: A housing nonprofit sees high average screening accuracy but unusually high rejection errors for one applicant group.
Explanation
Correct Answer
Compare relevant error outcomes across affected groups and investigate material disparities before deployment. Review group-level false-positive and false-negative rates with population context and document mitigation decisions. Together, they implement and test the intended safeguard. Aggregate accuracy can hide systematic harms to subpopulations even without explicit use of a protected attribute.
Incorrect Answers
This pair supports another governance intervention: Assess confidentiality, data minimization and adversarial misuse together in the AI risk register. For the actual case, the required action is: Compare relevant error outcomes across affected groups and investigate material disparities before deployment. The supporting evidence is: Review group-level false-positive and false-negative rates with population context and document mitigation decisions.
This pair supports another governance intervention: Classify a text-generating model separately from a numerical risk predictor when preparing evaluations. For the actual case, the required action is: Compare relevant error outcomes across affected groups and investigate material disparities before deployment. The supporting evidence is: Review group-level false-positive and false-negative rates with population context and document mitigation decisions.
This pair supports another governance intervention: Set an executive reporting cadence for AI inventory, residual risk and unresolved exceptions. For the actual case, the required action is: Compare relevant error outcomes across affected groups and investigate material disparities before deployment. The supporting evidence is: Review group-level false-positive and false-negative rates with population context and document mitigation decisions.
Question 36
Consider the following proposed AI use: A lending cooperative finds a strong overall credit score while certain geographic communities receive more false rejections. Which response best addresses the governance gap?
Explanation
Correct Answer
For the decision, compare relevant error outcomes across affected groups and investigate material disparities before deployment. Aggregate accuracy can hide systematic harms to subpopulations even without explicit use of a protected attribute. The practical acceptance evidence is: review group-level false-positive and false-negative rates with population context and document mitigation decisions
Incorrect Answers
For the decision, this would help address a separate governance task: Distinguish learned statistical prediction from a fixed rules engine before assigning AI-specific controls. For this case, however, aggregate accuracy can hide systematic harms to subpopulations even without explicit use of a protected attribute. The more direct step is: Compare relevant error outcomes across affected groups and investigate material disparities before deployment
This would help address a separate governance task: Classify a text-generating model separately from a numerical risk predictor when preparing evaluations. For this case, however, aggregate accuracy can hide systematic harms to subpopulations even without explicit use of a protected attribute. The more direct step is: Compare relevant error outcomes across affected groups and investigate material disparities before deployment.
This would help address a separate governance task: Record the system's intended purpose, risk assessments, decisions, changes and evidence in a controlled repository. For this case, however, aggregate accuracy can hide systematic harms to subpopulations even without explicit use of a protected attribute. The more direct step is: Compare relevant error outcomes across affected groups and investigate material disparities before deployment.
Question 37
The organization chooses to compare relevant error outcomes across affected groups and investigate material disparities before deployment for this situation: A lending cooperative finds a strong overall credit score while certain geographic communities receive more false rejections. Which rationale best supports that choice?
Explanation
Correct Answer
For the technical explanation, aggregate accuracy can hide systematic harms to subpopulations even without explicit use of a protected attribute. That is why the case calls for this action: compare relevant error outcomes across affected groups and investigate material disparities before deployment
Incorrect Answers
For the technical explanation, that explains why a team might classify a text-generating model separately from a numerical risk predictor when preparing evaluations, but it does not justify the decision required by the case. Aggregate accuracy can hide systematic harms to subpopulations even without explicit use of a protected attribute
For the technical explanation, that explains why a team might distinguish learned statistical prediction from a fixed rules engine before assigning AI-specific controls, but it does not justify the decision required by the case. Aggregate accuracy can hide systematic harms to subpopulations even without explicit use of a protected attribute
That explains why a team might set an executive reporting cadence for AI inventory, residual risk and unresolved exceptions, but it does not justify the decision required by the case. Aggregate accuracy can hide systematic harms to subpopulations even without explicit use of a protected attribute.
Question 38
The organization has implemented a control for this case: A lending cooperative finds a strong overall credit score while certain geographic communities receive more false rejections. Which result would best validate it?
Explanation
Correct Answer
For the verification, review group-level false-positive and false-negative rates with population context and document mitigation decisions. This verifies the necessary safeguard rather than merely showing that an AI project was completed. Aggregate accuracy can hide systematic harms to subpopulations even without explicit use of a protected attribute
Incorrect Answers
For the verification, test access controls, prompt abuse, output leakage and minimization against approved data-use boundaries would test a different outcome. The directly relevant evidence is to review group-level false-positive and false-negative rates with population context and document mitigation decisions
Document whether outputs come from fitted parameters or predefined rules, then map suitable tests to each component would test a different outcome. The directly relevant evidence is to review group-level false-positive and false-negative rates with population context and document mitigation decisions.
Sample incoming requests to confirm purpose, affected groups, data sources and an initial risk decision are documented would test a different outcome. The directly relevant evidence is to review group-level false-positive and false-negative rates with population context and document mitigation decisions.
Question 39
The organization has this problem: A lending cooperative finds a strong overall credit score while certain geographic communities receive more false rejections. It takes a shortcut: Approve a screening model solely because its overall accuracy exceeds the headline target. What is the principal risk?
Explanation
Correct Answer
For the risk comparison, a well-performing average can conceal severe errors concentrated in one affected group. The shortcut overlooks the central safeguard: compare relevant error outcomes across affected groups and investigate material disparities before deployment. Aggregate accuracy can hide systematic harms to subpopulations even without explicit use of a protected attribute
Incorrect Answers
This consequence follows more directly from another shortcut: Assume encrypted storage alone makes all model inputs and outputs privacy-safe. The shortcut actually described would instead create this risk: A well-performing average can conceal severe errors concentrated in one affected group.
For the risk comparison, this consequence follows more directly from another shortcut: Treat one satisfactory model response as proof that all future responses will be equally safe. The shortcut actually described would instead create this risk: A well-performing average can conceal severe errors concentrated in one affected group
This consequence follows more directly from another shortcut: Allow only the model builders to review a system that will affect diverse external users. The shortcut actually described would instead create this risk: A well-performing average can conceal severe errors concentrated in one affected group.
Question 40
The governance team needs both a decision and a way to verify it for this case: A lending cooperative finds a strong overall credit score while certain geographic communities receive more false rejections. Which pair is best?
Explanation
Correct Answer
For the action and verification, compare relevant error outcomes across affected groups and investigate material disparities before deployment. Review group-level false-positive and false-negative rates with population context and document mitigation decisions. Together, they implement and test the intended safeguard. Aggregate accuracy can hide systematic harms to subpopulations even without explicit use of a protected attribute
Incorrect Answers
This pair supports another governance intervention: Distinguish learned statistical prediction from a fixed rules engine before assigning AI-specific controls. For the actual case, the required action is: Compare relevant error outcomes across affected groups and investigate material disparities before deployment. The supporting evidence is: Review group-level false-positive and false-negative rates with population context and document mitigation decisions.
For the action and verification, this pair supports another governance intervention: Assess confidentiality, data minimization and adversarial misuse together in the AI risk register. For the actual case, the required action is: Compare relevant error outcomes across affected groups and investigate material disparities before deployment. The supporting evidence is: Review group-level false-positive and false-negative rates with population context and document mitigation decisions
This pair supports another governance intervention: Train developers, approvers, operators and frontline users on the distinct AI risks and decisions they control. For the actual case, the required action is: Compare relevant error outcomes across affected groups and investigate material disparities before deployment. The supporting evidence is: Review group-level false-positive and false-negative rates with population context and document mitigation decisions.
Artificial Intelligence Governance Professional (AIGP) Practice Tests
- IAPP AIGP Practice Test 2 - 10 Questions (Last Reviewed 20-Nov-2026)
- IAPP AIGP Practice Test 3 - 10 Questions (Last Reviewed 19-Nov-2026)
- IAPP AIGP Practice Test 4 - 10 Questions (Last Reviewed 18-Nov-2026)
- IAPP AIGP Practice Test 5 - 10 Questions (Last Reviewed 17-Nov-2026)
- IAPP AIGP Practice Test 6 - 10 Questions (Last Reviewed 16-Nov-2026)