IAPP AIGP Practice Test 23 - Prepaway
Practice Exams:

Pass IAPP AIGP Exam in First Attempt Guaranteed!

Get 100% Latest Exam Questions, Accurate & Verified Answers!

30 Days Free Updates, Instant Download!

Verified By Experts
Premium Bundle
$29.98
$79.99 $109.97

AIGP Premium Bundle


  • Premium File 222 Questions & Answers. Last update: Nov 15, 2026
  • Training Course 124 Lectures
  • Study Guide 566 Pages

IAPP AIGP Practice Test 23

Question 220

An architecture studio has no rule on whether employees may ask an AI tool to generate confidential client plans. Which test result would best demonstrate that the actual governance concern has been addressed?

  1. Contractual and technical evidence addresses lawful training data and allocated risks before procurement signoff.
  2. Design verification tests user recourse and affected-group outcomes alongside technical objectives.
  3. Sample workflows demonstrate confidential designs remain inside approved processing boundaries.
  4. Subcontractor changes trigger documented risk review and compliant data-handling checks.

Explanation

Correct Answer

C:

Sample workflows demonstrate confidential designs remain inside approved processing boundaries. Rules for employee AI usage must update existing confidentiality and data security controls.

Incorrect Answers

A:

Contractual and technical evidence addresses lawful training data and allocated risks before procurement signoff. Rules for employee AI usage must update existing confidentiality and data security controls. Sample workflows demonstrate confidential designs remain inside approved processing boundaries.

B:

Design verification tests user recourse and affected-group outcomes alongside technical goal. Rules for employee AI usage must update existing confidentiality and data security controls. Sample workflows demonstrate confidential designs remain inside approved processing boundaries.

D:

Subcontractor changes trigger documented risk review and compliant data-handling checks. Rules for employee AI usage must update existing confidentiality and data security controls. Sample workflows demonstrate confidential designs remain inside approved processing boundaries.

Question 221

A worldwide AI recruiting tool becomes restricted for a particular country, but administrators can only disable every market at once. Which additional observation would most strongly confirm the particular risk or control gap?

  1. Affected residents cannot correct their information or learn why assistance is delayed.
  2. Policy settings have no reliable geography or use-case restriction despite a regional compliance decision.
  3. Pilot audits show analysts treat a low-confidence output as a final mandatory decision.
  4. A read-only conversation has access to high-impact write operations through the same token.

Explanation

Correct Answer

B:

Policy settings have no reliable geography or use-case restriction despite a regional compliance decision.

Incorrect Answers

A:

Affected residents cannot correct their information or learn why assistance is delayed. Policy settings have no reliable geography or use-case restriction despite a regional compliance decision.

C:

Pilot audits show analysts treat a low-confidence output as a final mandatory decision. Policy settings have no reliable geography or use-case restriction despite a regional compliance decision.

D:

A read-only conversation has access to high-impact write operations through the same token. Policy settings have no reliable geography or use-case restriction despite a regional compliance decision.

Question 222

A customer-support AI routes uncertain legal questions to itself instead of human specialists because confidence is never checked. Which governance intervention most directly addresses this specific situation?

  1. Define uncertainty thresholds and human escalation for unsupported or high-stakes requests.
  2. Conduct an impact assessment including affected individuals, error severity and available recourse.
  3. Design representative benchmarks and controlled pilot tests before approving rollout.
  4. Include accessible alternative interaction paths in the design requirements and evaluate with affected users.

Explanation

Correct Answer

A:

Define uncertainty thresholds and human escalation for unsupported or high-stakes requests. The action addresses uncertainty threshold and fallback rather than treating every AI control as interchangeable. The decisive signal to verify is that low-confidence advice is sent automatically while specialists never receive the flagged conversation. This matters because model confidence and scope limits must influence system decisions, not merely appear in an appendix. Success should be measured by confirming that test cases outside supported knowledge reliably trigger the documented fallback process.

Incorrect Answers

B:

Conduct an impact assessment including affected individuals, error severity and available recourse. Model confidence and scope limits must influence system decisions, not merely appear in an appendix. Define uncertainty thresholds and human escalation for unsupported or high-stakes requests.

C:

Design representative benchmarks and controlled pilot tests before approving rollout. Model confidence and scope limits must influence system decisions, not merely appear in an appendix. Define uncertainty thresholds and human escalation for unsupported or high-stakes requests.

D:

Include accessible alternative interaction paths in the design requirements and evaluate with affected users. Model confidence and scope limits must influence system decisions, not merely appear in an appendix. Define uncertainty thresholds and human escalation for unsupported or high-stakes requests.

Question 223

A healthcare app stores only a final answer, losing the model version and prompt context needed to investigate a harmful suggestion. Which additional observation would most strongly confirm the particular risk or control gap?

  1. Investigators cannot reconstruct which configuration produced the reported advice.
  2. Different teams send client-sensitive plans to public assistants without a documented authorization standard.
  3. Configuration exports show indefinite logging with no retention owner or deletion route.
  4. A reviewer cannot connect datasets, test results, mitigations and the released version.

Explanation

Correct Answer

A:

Investigators cannot reconstruct which configuration produced the reported advice.

Incorrect Answers

B:

Different teams send client-sensitive plans to public assistants without a documented authorization standard. Investigators cannot reconstruct which configuration produced the reported advice.

C:

Configuration exports show indefinite logging with no retention owner or deletion route. Investigators cannot reconstruct which configuration produced the reported advice.

D:

A reviewer cannot connect datasets, test results, mitigations and the released version. Investigators cannot reconstruct which configuration produced the reported advice.

Question 224

A news organization plans to republish generated graphics but has not examined who bears infringement claims. Which governance intervention most directly addresses this specific situation?

  1. Review indemnity limits, exclusions and required user safeguards alongside licensing.
  2. Negotiate purpose limits, retention, deletion and training-use restrictions in the service agreement.
  3. Map all data flows including support and telemetry, then reconcile them with residency and transfer terms.
  4. Assess direct developer and deployer responsibilities, documentation and liability exposure.

Explanation

Correct Answer

A:

Review indemnity limits, exclusions and required user safeguards alongside licensing. The action addresses indemnification for third-party claims rather than treating every AI control as interchangeable. The decisive signal to verify is that the vendor offers a narrow warranty that excludes customer-provided reference images. This matters because contract risk allocation may depend on how the customer configures and uses the AI system. Success should be measured by confirming that legal approval records the actual allocation of claim risk for the publishing workflow.

Incorrect Answers

B:

Negotiate purpose limits, retention, deletion and training-use restrictions in the service agreement. Contract risk allocation may depend on how the customer configures and uses the AI system. Review indemnity limits, exclusions and required user safeguards alongside licensing.

C:

Map all data flows including support and telemetry, then reconcile them with residency and transfer terms. Contract risk allocation may depend on how the customer configures and uses the AI system. Review indemnity limits, exclusions and required user safeguards alongside licensing.

D:

Assess direct developer and deployer responsibilities, documentation and liability exposure. Contract risk allocation may depend on how the customer configures and uses the AI system. Review indemnity limits, exclusions and required user safeguards alongside licensing.

Question 225

A benefits provider drafts a welfare prediction feature based solely on throughput targets, leaving out appeal rights. Which governance reasoning best supports the appropriate response?

  1. AI supply-chain controls should follow material processing and contractual dependencies.
  2. Responsible AI requirements belong in the product design, not only in a final release memo.
  3. Rules for employee AI usage must update existing confidentiality and data security controls.
  4. An effective AI policy must be actionable when risks exceed tolerance.

Explanation

Correct Answer

B:

Responsible AI requirements belong in the product design, not only in a final release memo. This reasoning fits ethics-by-design check in specifications because the observed gap is that the product backlog includes speed metrics but no user-harm or review criteria.

Incorrect Answers

A:

AI supply-chain controls should follow material processing and contractual dependencies. This reasoning fits ethics-by-design check in specifications because the observed gap is that the product backlog includes speed metrics but no user-harm or review criteria. Responsible AI requirements belong in the product design, not only in a final release memo.

C:

Rules for employee AI usage must update existing confidentiality and data security controls. This reasoning fits ethics-by-design check in specifications because the observed gap is that the product backlog includes speed metrics but no user-harm or review criteria. Responsible AI requirements belong in the product design, not only in a final release memo.

D:

An effective AI policy must be actionable when risks exceed tolerance. This reasoning fits ethics-by-design check in specifications because the observed gap is that the product backlog includes speed metrics but no user-harm or review criteria. Responsible AI requirements belong in the product design, not only in a final release memo.

Question 226

A retail forecast model abruptly degrades after a data pipeline changes units from dollars to cents. Select the ONE intervention-and-verification pair that directly resolves this specific governance gap.

  1. Intervention: Version code, dependencies, datasets and evaluation settings for reproducible experiments. / Verification: Invalid batches are blocked and recovery replays the last accepted data version.
  2. Intervention: Measure calibration alongside discrimination and correct or qualify probability estimates. / Verification: Post-validation probabilities are calibrated for the intended decision context or clearly limited.
  3. Intervention: Add schema, range and distribution validation at ingestion with lineage and rollback. / Verification: Validated privacy tests support the release conditions and risky synthetic records are excluded.
  4. Intervention: Add schema, range and distribution validation at ingestion with lineage and rollback. / Verification: Invalid batches are blocked and recovery replays the last accepted data version.

Explanation

Correct Answer

D:

Intervention: Add schema, range and distribution validation at ingestion with lineage and rollback. / Verification: Invalid batches are blocked and recovery replays the last accepted data version. The case is signaled by the fact that feature values jump by a factor of 100 without a corresponding real business change.

Incorrect Answers

A:

Intervention: Version code, dependencies, datasets and evaluation settings for reproducible experiments. / Verification: Invalid batches are blocked and recovery replays the last accepted data version. The case is signaled by the fact that feature values jump by a factor of 100 without a corresponding real business change. Intervention: Add schema, range and distribution validation at ingestion with lineage and rollback. / Verification: Invalid batches are blocked and recovery replays the last accepted data version.

B:

Intervention: Measure calibration alongside discrimination and correct or qualify probability estimates. / Verification: Post-validation probabilities are calibrated for the intended decision context or clearly limited. The case is signaled by the fact that feature values jump by a factor of 100 without a corresponding real business change. Intervention: Add schema, range and distribution validation at ingestion with lineage and rollback. / Verification: Invalid batches are blocked and recovery replays the last accepted data version.

C:

Intervention: Add schema, range and distribution validation at ingestion with lineage and rollback. / Verification: Validated privacy tests support the release conditions and risky synthetic records are excluded. The case is signaled by the fact that feature values jump by a factor of 100 without a corresponding real business change. Intervention: Add schema, range and distribution validation at ingestion with lineage and rollback. / Verification: Invalid batches are blocked and recovery replays the last accepted data version.

Question 227

A fintech company ships an updated anti-fraud model despite unresolved severe subgroup false positives. Which governance intervention most directly addresses this specific situation?

  1. Apply a formal release gate with tolerances, remediation requirements and stop authority.
  2. Threat-model the artifact, build pipeline and inference dependencies and add integrity controls.
  3. Monitor subgroup outcomes after launch and route significant discrepancies to accountable review.
  4. Set harm-specific monitoring thresholds and an alert-to-containment workflow.

Explanation

Correct Answer

A:

Apply a formal release gate with tolerances, remediation requirements and stop authority. The action addresses go-no-go release criteria rather than treating every AI control as interchangeable. The decisive signal to verify is that validation shows an unresolved high-severity error beyond the agreed acceptance threshold. This matters because the existence of a trained model does not by itself justify deployment approval. Success should be measured by confirming that only releases meeting documented safety, fairness and performance gates enter production.

Incorrect Answers

B:

Threat-model the artifact, build pipeline and inference dependencies and add integrity controls. The existence of a trained model does not by itself justify deployment approval. Apply a formal release gate with tolerances, remediation requirements and stop authority.

C:

Monitor subgroup outcomes after launch and route significant discrepancies to accountable review. The existence of a trained model does not by itself justify deployment approval. Apply a formal release gate with tolerances, remediation requirements and stop authority.

D:

Set harm-specific monitoring thresholds and an alert-to-containment workflow. The existence of a trained model does not by itself justify deployment approval. Apply a formal release gate with tolerances, remediation requirements and stop authority.

Question 228

A compliance team chooses an expensive generative model to label three fixed document categories without evaluating a simpler supervised classifier. Which test result would best demonstrate that the actual governance concern has been addressed?

  1. Adversarial inputs are quarantined or rejected before they affect approved training assets.
  2. A documented override route prevents contested cancellations without appropriate authorization.
  3. The design review documents input and mitigation decisions for each materially affected group.
  4. An evaluation chooses the simplest suitable model meeting performance and governance goals.

Explanation

Correct Answer

D:

An evaluation chooses the simplest suitable model meeting performance and governance goals. Architecture decisions should be driven by fit-for-purpose evidence rather than model popularity.

Incorrect Answers

A:

Adversarial inputs are quarantined or rejected before they affect approved training assets. Architecture decisions should be driven by fit-for-purpose evidence rather than model popularity. An evaluation chooses the simplest suitable model meeting performance and governance goals.

B:

A documented override route prevents contested cancellations without appropriate authorization. Architecture decisions should be driven by fit-for-purpose evidence rather than model popularity. An evaluation chooses the simplest suitable model meeting performance and governance goals.

C:

The design review documents input and mitigation decisions for each materially affected group. Architecture decisions should be driven by fit-for-purpose evidence rather than model popularity. An evaluation chooses the simplest suitable model meeting performance and governance goals.

Question 229

An insurer discovers the newest claims model causes large unexplained rejection spikes but no tested previous version is preserved. Which governance reasoning best supports the appropriate response?

  1. Periodic assurance must challenge the conditions and metrics used to declare a model reliable.
  2. The existence of a trained model does not by itself justify deployment approval.
  3. Safe AI release management includes the ability to reverse harmful changes.
  4. Third-party AI changes create a release risk even when the integrating app code is unchanged.

Explanation

Correct Answer

C:

Safe AI release management includes the ability to reverse harmful changes. This reasoning fits rollback of harmful version because the observed gap is that operations has only the latest model image and no prior compatible feature schema. It points to the practical intervention: Maintain validated rollback artifacts and a rehearsed controlled reversion procedure. The outcome should then show that a drill restores a previously approved model and preserves audit evidence without data loss.

Incorrect Answers

A:

Periodic assurance must challenge the conditions and metrics used to declare a model reliable. This reasoning fits rollback of harmful version because the observed gap is that operations has only the latest model image and no prior compatible feature schema. Safe AI release management includes the ability to reverse harmful changes.

B:

The existence of a trained model does not by itself justify deployment approval. This reasoning fits rollback of harmful version because the observed gap is that operations has only the latest model image and no prior compatible feature schema. Safe AI release management includes the ability to reverse harmful changes.

D:

Third-party AI changes create a release risk even when the integrating app code is unchanged. This reasoning fits rollback of harmful version because the observed gap is that operations has only the latest model image and no prior compatible feature schema. Safe AI release management includes the ability to reverse harmful changes.

Question 230

A bank specifies only overall fraud-detection accuracy for a model affecting customers from several demographic groups. Which test result would best demonstrate that the actual governance concern has been addressed?

  1. Testing shows the design removes or reduces the hazard at source, with warnings supplementary.
  2. Threat and access tests demonstrate employee information is restricted to authorized purposes.
  3. Signed acceptance tests cover relevant group performance and meaningful review of flagged accounts.
  4. An audit can trace every consequential feature and mitigation to a reviewable design decision.

Explanation

Correct Answer

C:

Signed acceptance tests cover relevant group performance and meaningful review of flagged accounts. The required intervention is to add subgroup error thresholds and a remediation plan to the design acceptance criteria. The particular diagnostic issue is that the requirements document hides false-positive differences by reporting one aggregate statistic. The governing distinction is that fairness is a measurable design requirement where unequal errors affect people.

Incorrect Answers

A:

Testing shows the design removes or reduces the hazard at source, with warnings supplementary. Fairness is a measurable design requirement where unequal errors affect people. Signed acceptance tests cover relevant group performance and meaningful review of flagged accounts.

B:

Threat and access tests demonstrate employee information is restricted to authorized purposes. Fairness is a measurable design requirement where unequal errors affect people. Signed acceptance tests cover relevant group performance and meaningful review of flagged accounts.

D:

An audit can trace every consequential feature and mitigation to a reviewable design decision. Fairness is a measurable design requirement where unequal errors affect people. Signed acceptance tests cover relevant group performance and meaningful review of flagged accounts.

Artificial Intelligence Governance Professional (AIGP) Practice Tests