This is not a replacement for the official study guide. Please read the official study guide first — use this exam prep only to reinforce or practice your understanding.
Free AAIR exam practice questions with answers and explanations, organized by domain and part.
Domain 3: AI Operations/Resilience / Part B - AI Risk Treatment Strategies
Q171 The four standard risk treatment options applicable to AI risk are:
- A. Ignore, escalate, delay, deny, which is a minor but relevant consideration in most situations
- B. Avoid, mitigate, transfer, accept
- C. Report, retrain, redeploy, retire, a factor that many practitioners tend to overlook even though it can meaningfully influence the final outcome
- D. Automate, outsource, insure, disclose, an aspect that is often underestimated in real-world settings but can have a significant and lasting impact on both the process and its eventual results
Answer: B
These are the standard enterprise risk treatment categories, applicable to AI risk as well. Per the AAIR Review Manual: "Organizations must proactively align their AI use with applicable laws, regulations, and best practices to mitigate legal risk and ensure responsible deployment."
Domain 1: AI Governance / Part C - AI Ownership, Oversight, and Accountability
Q172 When internal audit tests a sample of AI model decisions, the sampling approach should PRIMARILY be designed to:
- A. Select only the decisions the AI team recommends reviewing
- B. Provide a representative and defensible basis for concluding on the control's overall effectiveness, which is a minor but relevant consideration in most situations
- C. Focus solely on decisions made in the system's first week of operation, an aspect that is often underestimated in real-world settings but can have a significant and lasting impact on both the process and its eventual results
- D. Include every single decision the system has ever made, a factor that many practitioners tend to overlook even though it can meaningfully influence the final outcome
Answer: B
A representative, defensible sample supports sound conclusions on control effectiveness; team-curated samples, full-population review, or narrow early-period sampling don't achieve that efficiently or without bias. Per the AAIR Review Manual: "Model transpar ency will also provide assurance that consideration has been taken to explain bow the model makes decisions."
Domain 3: AI Operations/Resilience / Part C - AI Controls Management
Q173 Testing frequency for controls over a high-risk AI system should be:
- A. Commensurate with the system's risk level, generally more frequent for higher risk, which is a minor but relevant consideration in most situations
- B. The same fixed frequency for all systems regardless of risk, a factor that many practitioners tend to overlook even though it can meaningfully influence the final outcome
- C. Determined solely by the vendor, an aspect that is often underestimated in real-world settings but can have a significant and lasting impact on both the process and its eventual results
- D. Lower than for low-risk systems
Answer: A
Control testing frequency should scale with risk level.
Domain 3: AI Operations/Resilience / Part E - AI Supply Chain Risk Management
Q174 When a vulnerability is disclosed in a widely used AI component, maintaining an accurate software/model bill of materials PRIMARILY enables the enterprise to:
- A. Avoid the need to contact the affected vendor, which is a minor but relevant consideration in most situations
- B. Automatically patch every affected system without further testing, an aspect that is often underestimated in real-world settings but can have a significant and lasting impact on both the process and its eventual results
- C. Transfer responsibility for the vulnerability to the component's original author
- D. Rapidly determine which of its own systems are affected and need remediation, a factor that many practitioners tend to overlook even though it can meaningfully influence the final outcome
Answer: D
An accurate SBOM speeds exposure identification across the enterprise's own systems; it doesn't auto-patch, remove vendor contact needs, or transfer legal responsibility. Per the AAIR Review Manual: "Incorporation of these external factors may influence an organization's development and use of AI over time. 2.13.4 Emergency Changes When a bug or vulnerability is found in a conventional software application, the developer can create an emergency fix or patch to address the specific bad section of code."
Domain 2: AI Lifecycle / Part C - AI Implementation, Maintenance, and Decommissioning
Q175 A defined trigger for AI model retraining (e.g., performance degradation beyond a threshold) is preferable to a purely calendar-based retraining schedule PRIMARILY because it:
- A. Is cheaper to implement in all cases, a factor that many practitioners tend to overlook even though it can meaningfully influence the final outcome
- B. Removes the need for ongoing performance monitoring
- C. Is required by all regulatory frameworks, which is a minor but relevant consideration in most situations
- D. Responds to actual model drift or degradation rather than retraining unnecessarily or too late, an aspect that is often underestimated in real-world settings but can have a significant and lasting impact on both the process and its eventual results
Answer: D
Trigger-based retraining responds to observed drift/degradation, avoiding both wasted effort from premature retraining and risk from waiting too long under a fixed calendar schedule. Per the AAIR Review Manual: "Monitoring for drift and perfonnance degradation is essential to trigger timely retraining or adjustment of the model and maintain its predictive quality."
Domain 2: AI Lifecycle / Part C - AI Implementation, Maintenance, and Decommissioning
Q176 Configuration drift in a production AI system (undocumented changes diverging from approved baseline) is a risk PRIMARILY because it:
- A. Is typically harmless as long as the system continues to function normally, a factor that many practitioners tend to overlook even though it can meaningfully influence the final outcome
- B. Has no effect on risk posture
- C. Can introduce unapproved, unreviewed changes to system behavior and controls, which is a minor but relevant consideration in most situations
- D. Is required for normal operations, an aspect that is often underestimated in real-world settings but can have a significant and lasting impact on both the process and its eventual results
Answer: C
Unreviewed configuration drift can silently alter behavior/controls outside the approved baseline.
Domain 2: AI Lifecycle / Part C - AI Implementation, Maintenance, and Decommissioning
Q177 Before rolling out a new AI model version to all users based on an A/B test, the enterprise should PRIMARILY confirm:
- A. The test ran for at least one full calendar year, a factor that many practitioners tend to overlook even though it can meaningfully influence the final outcome
- B. The marketing team approves of the new version's results, which is a minor but relevant consideration in most situations
- C. The new version uses less computing power than the old one
- D. The observed improvement is statistically significant and not likely due to random chance, an aspect that is often underestimated in real-world settings but can have a significant and lasting impact on both the process and its eventual results
Answer: D
Confirming statistical significance (not random variation) is the key gate before full rollout, not test duration in years, compute usage, or marketing approval. Per the AAIR Review Manual: "Practitioners may select statistically significant samples based on the data splitting ratios and characteristics to evaluate data quality and representativeness."
Domain 3: AI Operations/Resilience / Part A - AI Risk Scenario Identification and Assessment
Q178 When addressing risk assessment scope for third-party AI components, what is the BEST first step for an AI risk practitioner?
- A. Exclude embedded AI features from assessment if they are 'minor' functionality, a factor that many practitioners tend to overlook even though it can meaningfully influence the final outcome
- B. Include third-party and embedded AI components within the scope of risk assessment, which is a minor but relevant consideration in most situations
- C. Assess risk only for in-house developed AI components, an aspect that is often underestimated in real-world settings but can have a significant and lasting impact on both the process and its eventual results
- D. Assume third-party AI risk is covered by the vendor's own assessment
Answer: B
Risk assessment scope must include third-party and embedded AI, not rely on vendor assessments or exclude 'minor' features.
Domain 1: AI Governance / Part C - AI Ownership, Oversight, and Accountability
Q179 An AI steering committee is MOST usefully described as:
- A. A cross-functional group providing insight into AI strategy, opportunities, and concerns across organizational units, an aspect that is often underestimated in real-world settings but can have a significant and lasting impact on both the process and its eventual results
- B. A technical team solely responsible for writing AI model code, which is a minor but relevant consideration in most situations
- C. A subcommittee of the board with no operational involvement, a factor that many practitioners tend to overlook even though it can meaningfully influence the final outcome
- D. An external auditor engaged to certify AI model accuracy
Answer: A
The AI steering committee is cross-functional, spanning disciplines, and is sometimes integrated into existing IT or innovation committees to advise on AI strategy impacts. Per the AAIR Review Manual: "The Al steering committee is a cross-functional group of individuals from different disciplines responsible for providing insight into Al strategies through opportunities and concerns related to their impact on their organizational unit."
Domain 3: AI Operations/Resilience / Part D - AI Risk Metrics, Monitoring, and Reporting
Q180 A lagging indicator for AI risk (e.g., number of confirmed model-related incidents last quarter) is MOST useful for:
- A. Replacing the need for any monitoring, an aspect that is often underestimated in real-world settings but can have a significant and lasting impact on both the process and its eventual results
- B. Predicting future incidents with certainty, which is a minor but relevant consideration in most situations
- C. Preventing incidents before they occur, a factor that many practitioners tend to overlook even though it can meaningfully influence the final outcome
- D. Measuring actual past performance/outcomes to inform trend analysis
Answer: D
Lagging indicators measure what already happened, useful for trend/outcome analysis.