Pass IAPP AIGP Exam in First Attempt Guaranteed!
Get 100% Latest Exam Questions, Accurate & Verified Answers!
30 Days Free Updates, Instant Download!
IAPP AIGP Practice Test 10
Question 90
A disaster-response risk model outputs probabilities, yet events predicted at 90 percent occur only half the time. Select the ONE intervention-and-verification pair that directly resolves this specific governance gap.
Explanation
Correct Answer
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.
Incorrect Answers
Intervention: Measure calibration alongside discrimination and correct or qualify probability estimates. / Verification: The training manifest includes authorized sources with documented legal basis and exclusions. 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.
Intervention: Stratify and validate training and test samples for the actual served language communities. / Verification: Post-validation probabilities are calibrated for the intended decision context or clearly limited. 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.
Intervention: Split and audit data at the appropriate individual level to prevent leakage. / Verification: A leakage-free holdout evaluation reports realistic generalization to unseen applicants. 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.
Question 91
A legal-aid nonprofit pilots text generation that cites nonexistent local regulations as grounds to deny assistance. Select the ONE intervention-and-verification pair that directly resolves this specific governance gap.
Explanation
Correct Answer
Intervention: Require source verification, constrained outputs and escalation instead of treating generated citations as authority. / Verification: All asserted legal citations in acceptance tests resolve to approved sources or are withheld for review. The reason to require this combination is that fluent generative text can be unsupported; truthfulness demands verification rather than stylistic confidence.
Incorrect Answers
Intervention: Examine subgroup outcomes with validated fairness metrics and remediate job-irrelevant disparities before expansion. / Verification: Subgroup checks and documented remediation meet the approved nondiscrimination thresholds. The reason to require this combination is that fluent generative text can be unsupported; truthfulness demands verification rather than stylistic confidence. Intervention: Require source verification, constrained outputs and escalation instead of treating generated citations as authority. / Verification: All asserted legal citations in acceptance tests resolve to approved sources or are withheld for review.
Intervention: Require source verification, constrained outputs and escalation instead of treating generated citations as authority. / Verification: Test queries that lack supporting policy lead to an uncertainty response or human handoff rather than an invented ruling. The reason to require this combination is that fluent generative text can be unsupported; truthfulness demands verification rather than stylistic confidence. Intervention: Require source verification, constrained outputs and escalation instead of treating generated citations as authority. / Verification: All asserted legal citations in acceptance tests resolve to approved sources or are withheld for review.
Intervention: Evaluate coverage with representative dialect examples and plan targeted data and process remediation. / Verification: All asserted legal citations in acceptance tests resolve to approved sources or are withheld for review. The reason to require this combination is that fluent generative text can be unsupported; truthfulness demands verification rather than stylistic confidence. Intervention: Require source verification, constrained outputs and escalation instead of treating generated citations as authority. / Verification: All asserted legal citations in acceptance tests resolve to approved sources or are withheld for review.
Question 92
An insurer reports excellent risk prediction, but records from the same applicant appear in both training and evaluation datasets. Which additional observation would most strongly confirm the particular risk or control gap?
Explanation
Correct Answer
Identity matches show related records from one person in both train and test partitions.
Incorrect Answers
The same borderline content receives inconsistent labels across annotators without resolution. Identity matches show related records from one person in both train and test partitions.
Reliability diagrams show severe mismatch between forecast bands and observed frequency. Identity matches show related records from one person in both train and test partitions.
The team cannot reproduce how the feature used in a disputed decision was calculated. Identity matches show related records from one person in both train and test partitions.
Question 93
A public health authority hires a cloud vendor to process survey text but leaves instructions and audit rights undefined. Which additional observation would most strongly confirm the particular risk or control gap?
Explanation
Correct Answer
The draft contract describes infrastructure support but not processing limits or audit cooperation.
Incorrect Answers
The proposal stores face embeddings linked to student identities without special-category analysis. The draft contract describes infrastructure support but not processing limits or audit cooperation.
Prototype searches return another partner's restricted litigation documents to an unrelated employee. The draft contract describes infrastructure support but not processing limits or audit cooperation.
The source file is corrected while the deployed feature store still contains the inaccurate old field. The draft contract describes infrastructure support but not processing limits or audit cooperation.
Question 94
A benefits agency deploys fraud screening without a simple channel for beneficiaries to dispute incorrect flags. Which governance reasoning best supports the appropriate response?
Explanation
Correct Answer
Contestability helps protect individuals when AI informs consequential decisions. This reasoning fits complaint and appeal route because the observed gap is that affected residents cannot correct their information or learn why assistance is delayed. It points to the practical intervention: Establish an accessible appeal and human case review route with recorded outcomes. The outcome should then show that a test appeal reaches an authorized reviewer and resolves documented factual errors.
Incorrect Answers
Post-deployment data shifts require continuous monitoring and proportionate response. This reasoning fits complaint and appeal route because the observed gap is that affected residents cannot correct their information or learn why assistance is delayed. Contestability helps protect individuals when AI informs consequential decisions.
AI deactivation and localization controls should match the scope of legal and operational risk. This reasoning fits complaint and appeal route because the observed gap is that affected residents cannot correct their information or learn why assistance is delayed. Contestability helps protect individuals when AI informs consequential decisions.
AI authorization controls must follow real identity and role changes. This reasoning fits complaint and appeal route because the observed gap is that affected residents cannot correct their information or learn why assistance is delayed. Contestability helps protect individuals when AI informs consequential decisions.
Question 95
An online lender plans to infer financial distress from user activity but labels the processing compliant merely because a vendor has a privacy policy. Which governance reasoning best supports the appropriate response?
Explanation
Correct Answer
A supplier's privacy notice does not establish the customer controller's own legal grounds. This reasoning fits lawful basis for predictive profiling because the observed gap is that the vendor notice exists, but the lender has no record explaining the legal basis for its distinct profiling activity. It points to the practical intervention: Identify and document the controller's applicable lawful basis and associated fairness safeguards. The outcome should then show that a reviewed processing record establishes a suitable legal basis and necessary individual protections for this use.
Incorrect Answers
Outsourcing computation does not remove the controller's data-protection accountability. This reasoning fits lawful basis for predictive profiling because the observed gap is that the vendor notice exists, but the lender has no record explaining the legal basis for its distinct profiling activity. A supplier's privacy notice does not establish the customer controller's own legal grounds.
Data-subject rights require attention to AI-related copies and downstream processing, not just front-end records. This reasoning fits lawful basis for predictive profiling because the observed gap is that the vendor notice exists, but the lender has no record explaining the legal basis for its distinct profiling activity. A supplier's privacy notice does not establish the customer controller's own legal grounds.
Biometric identification can raise heightened privacy risks and legal protection needs. This reasoning fits lawful basis for predictive profiling because the observed gap is that the vendor notice exists, but the lender has no record explaining the legal basis for its distinct profiling activity. A supplier's privacy notice does not establish the customer controller's own legal grounds.
Question 96
A hiring-screening model has strong overall accuracy yet rejects qualified applicants in one protected group at a higher rate. Which test result would best demonstrate that the actual governance concern has been addressed?
Explanation
Correct Answer
Subgroup checks and documented remediation meet the approved nondiscrimination thresholds. Aggregate accuracy can hide discriminatory patterns and must be assessed against affected groups.
Incorrect Answers
Test queries that lack supporting policy lead to an uncertainty response or human handoff rather than an invented ruling. Aggregate accuracy can hide discriminatory patterns and must be assessed against affected groups. Subgroup checks and documented remediation meet the approved nondiscrimination thresholds.
Pilot prompts exclude unnecessary identifiers and approved retention tests verify protected handling. Aggregate accuracy can hide discriminatory patterns and must be assessed against affected groups. Subgroup checks and documented remediation meet the approved nondiscrimination thresholds.
Students can request review, correct relevant facts and receive a recorded decision from an authorized reviewer. Aggregate accuracy can hide discriminatory patterns and must be assessed against affected groups. Subgroup checks and documented remediation meet the approved nondiscrimination thresholds.
Question 97
A publisher finds an AI startup scraped subscription articles for training despite license terms restricting automated reuse. Which governance intervention most directly addresses this specific situation?
Explanation
Correct Answer
Assess copyright and license permissions and remove or remediate unauthorized training use. The action addresses licensed training content rather than treating every AI control as interchangeable. The decisive signal to verify is that dataset manifests identify paid articles whose licenses exclude the proposed training purpose. This matters because public accessibility does not necessarily grant a license for every AI training use. Success should be measured by confirming that the training-data register contains documented rights or excludes nonpermitted works.
Incorrect Answers
Assess applicable employment nondiscrimination duties and validate the model for unlawful disparate outcomes. Public accessibility does not necessarily grant a license for every AI training use. Assess copyright and license permissions and remove or remediate unauthorized training use.
Review source-license restrictions and output rights before commercial distribution. Public accessibility does not necessarily grant a license for every AI training use. Assess copyright and license permissions and remove or remediate unauthorized training use.
Assess housing discrimination risk and remove unjustified proxy effects from the screening process. Public accessibility does not necessarily grant a license for every AI training use. Assess copyright and license permissions and remove or remediate unauthorized training use.
Question 98
A railway maintenance classifier reaches 98 percent accuracy because almost all examples are normal, while rare failures go undetected. Which governance reasoning best supports the appropriate response?
Explanation
Correct Answer
Evaluation metrics must match the consequences of errors and event prevalence. This reasoning fits rare-event evaluation for safety incidents because the observed gap is that the aggregate metric is dominated by normal cases and masks missed hazards. It points to the practical intervention: Evaluate recall, false negatives and relevant operating thresholds for rare safety events. The outcome should then show that independent rare-event tests meet safety-relevant detection criteria with documented limits.
Incorrect Answers
Training distribution should represent the deployment population rather than the easiest data source. This reasoning fits rare-event evaluation for safety incidents because the observed gap is that the aggregate metric is dominated by normal cases and masks missed hazards. Evaluation metrics must match the consequences of errors and event prevalence.
Data integrity checks are essential governance controls for model training and retraining. This reasoning fits rare-event evaluation for safety incidents because the observed gap is that the aggregate metric is dominated by normal cases and masks missed hazards. Evaluation metrics must match the consequences of errors and event prevalence.
Overlapping training and test information can produce falsely optimistic performance claims. This reasoning fits rare-event evaluation for safety incidents because the observed gap is that the aggregate metric is dominated by normal cases and masks missed hazards. Evaluation metrics must match the consequences of errors and event prevalence.
Question 99
A customer kiosk needs instant short answers, while the selected model consumes extreme compute for marginal quality improvement. Which governance reasoning best supports the appropriate response?
Explanation
Correct Answer
A fit-for-purpose AI deployment balances efficacy with performance and sustainability. This reasoning fits inference latency and energy tradeoff because the observed gap is that trials show a large model misses interaction deadlines for a small accuracy gain. It points to the practical intervention: Benchmark acceptable accuracy against cost, latency and energy use rather than optimizing one measure. The outcome should then show that the selected deployment meets response, quality and operating budgets together.
Incorrect Answers
Operational adoption requires prepared human users as well as working software. This reasoning fits inference latency and energy tradeoff because the observed gap is that trials show a large model misses interaction deadlines for a small accuracy gain. A fit-for-purpose AI deployment balances efficacy with performance and sustainability.
Model adaptation choices require evidence of task fit, data suitability and regression risk. This reasoning fits inference latency and energy tradeoff because the observed gap is that trials show a large model misses interaction deadlines for a small accuracy gain. A fit-for-purpose AI deployment balances efficacy with performance and sustainability.
Deployment placement should reflect latency, connectivity, safety and maintenance risks. This reasoning fits inference latency and energy tradeoff because the observed gap is that trials show a large model misses interaction deadlines for a small accuracy gain. A fit-for-purpose AI deployment balances efficacy with performance and sustainability.
Question 100
An HR analytics tool built to summarize workforce training is reused to rank employees for dismissal. Which governance intervention most directly addresses this specific situation?
Explanation
Correct Answer
Monitor and restrict secondary uses outside the assessed purpose, with new impact and legal review. The action addresses downstream repurposing by employees rather than treating every AI control as interchangeable. The decisive signal to verify is that audit logs show training summaries exported into a disciplinary scoring workflow. This matters because a benign original use does not justify materially different consequential downstream decisions. Success should be measured by confirming that the unapproved personnel ranking stops and any new use requires separate authorization.
Incorrect Answers
Create a safe human fallback with adequate staffing, queues and incident escalation. A benign original use does not justify materially different consequential downstream decisions. Monitor and restrict secondary uses outside the assessed purpose, with new impact and legal review.
Require release comparison, rollback readiness and monitored validation of production updates. A benign original use does not justify materially different consequential downstream decisions. Monitor and restrict secondary uses outside the assessed purpose, with new impact and legal review.
Prepare a coordinated, truthful external communication plan with responsible spokespeople. A benign original use does not justify materially different consequential downstream decisions. Monitor and restrict secondary uses outside the assessed purpose, with new impact and legal review.
Artificial Intelligence Governance Professional (AIGP) Practice Tests
- IAPP AIGP Practice Test 8 - 10 Questions (Last Reviewed 14-Nov-2026)
- IAPP AIGP Practice Test 9 - 10 Questions (Last Reviewed 13-Nov-2026)
- IAPP AIGP Practice Test 10 - 10 Questions (Last Reviewed 12-Nov-2026)
- IAPP AIGP Practice Test 11 - 10 Questions (Last Reviewed 11-Nov-2026)
- IAPP AIGP Practice Test 12 - 10 Questions (Last Reviewed 10-Nov-2026)
