Free AAIR exam practice questions with answers and explanations, organized by domain and part.
Domain 2: AI Lifecycle / Part B - AI Model Training, Testing, and Validation
Q161 In the context of independent validation of model results, which of the following represents sound AI risk management?
- A. Have model validation performed by a function independent from model development
- B. Allow the development team to self-validate without independent review
- C. Use independent validation only for regulator-mandated models
- D. Skip independent validation for internally facing tools
Answer: A
Independent validation reduces conflict-of-interest risk and should not be limited to regulated or self-validated cases.
Domain 1: AI Governance / Part B - AI Organizational Processes and Alignment
Q162 In the context of alignment of AI metrics with business KPIs, which of the following represents sound AI risk management?
- A. Tie AI performance metrics back to relevant business KPIs and risk indicators
- B. Track AI metrics in isolation from business performance
- C. Only track technical accuracy metrics
- D. Avoid defining AI metrics until problems arise
Answer: A
AI metrics should connect to business KPIs and risk indicators, not remain purely technical or reactive.
Domain 2: AI Lifecycle / Part A - AI Design, Development/Procurement, and Documentation
Q163 Regarding AI system design requirements traceability, the BEST practice is to:
- A. Maintain traceability between business requirements and AI system design decisions
- B. Design AI systems without documenting requirement linkage
- C. Trace requirements only for externally audited systems
- D. Document traceability only after deployment issues arise
Answer: A
Design traceability to requirements should be maintained throughout, not skipped or added reactively.
Domain 2: AI Lifecycle / Part D - AI Data and Asset Management
Q164 Maintaining an inventory of all AI models in use is MOST important for:
- A. Enabling risk oversight, including identifying shadow AI and unmanaged exposure
- B. Satisfying a one-time audit only
- C. Reducing the number of models built
- D. Replacing policy documentation
Answer: A
An asset inventory is foundational for identifying unmanaged ("shadow") AI risk.
Domain 3: AI Operations/Resilience / Part F - AI Incident Response, BIA, Business Continuity, and Disaster Recovery
Q165 Regarding AI incident classification scheme, the BEST practice is to:
- A. Limit incident classification to security-related AI incidents
- B. Classify AI incidents by severity and type to drive proportionate response actions
- C. Treat all AI incidents with an identical response regardless of severity
- D. Classify incidents only after the response has already concluded
Answer: B
Severity/type-based classification enables proportionate response, rather than uniform, after-the-fact, or security-only classification.
Domain 1: AI Governance / Part F - AI Trustworthiness, Ethical, and Societal Implications
Q166 A disparate-impact analysis on AI model outcomes is used to detect:
- A. Hardware inefficiency
- B. Unequal outcomes across protected groups that may indicate bias
- C. Network latency issues
- D. Vendor pricing discrepancies
Answer: B
Disparate-impact analysis specifically targets outcome inequities across protected groups.
Domain 1: AI Governance / Part F - AI Trustworthiness, Ethical, and Societal Implications
Q167 "Explainability" in AI trustworthiness refers PRIMARILY to:
- A. The model's processing speed
- B. The ability to describe how a model reached a given output in understandable terms
- C. The cost of running the model
- D. The model's popularity
Answer: B
Explainability concerns understandable insight into how outputs are derived.
Domain 2: AI Lifecycle / Part C - AI Implementation, Maintenance, and Decommissioning
Q168 A/B testing a new AI model version against the existing production model with live traffic splits is MOST useful for:
- A. Satisfying regulatory filing requirements only
- B. Permanently replacing the need for any pre-production testing
- C. Avoiding the need for a rollback plan
- D. Comparing real-world performance of both versions under equivalent conditions before committing to a full cutover
Answer: D
A/B testing provides a controlled, real-world comparison between model versions, informing the cutover decision; it supplements rather than replaces pre-production testing and rollback planning.
Domain 3: AI Operations/Resilience / Part D - AI Risk Metrics, Monitoring, and Reporting
Q169 A model performance monitoring dashboard is MOST useful for:
- A. Replacing the need for periodic model validation
- B. Giving stakeholders ongoing visibility into model behavior and emerging issues
- C. Satisfying only aesthetic reporting preferences
- D. Eliminating the need for KRIs
Answer: B
Dashboards provide ongoing visibility but complement, not replace, periodic validation.
Domain 2: AI Lifecycle / Part C - AI Implementation, Maintenance, and Decommissioning
Q170 Before decommissioning an AI system, the enterprise should PRIMARILY ensure:
- A. Immediate deletion of all related data with no review
- B. Dependent processes are identified and data/retention obligations are addressed
- C. No documentation is required since the system is being removed
- D. The vendor is not notified
Answer: B
Decommissioning requires addressing dependencies and retention obligations, not abrupt deletion.