Free AAIR exam practice questions with answers and explanations, organized by domain and part.
Domain 1: AI Governance / Part E - AI Regulatory Compliance and Legal Considerations
Q481 When addressing regulatory horizon scanning for AI, what is the BEST first step for an AI risk practitioner?
- A. Review AI regulatory developments only once a year, 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. Rely on vendors to notify the enterprise of regulatory changes
- C. Maintain an ongoing process to monitor emerging AI regulation relevant to the enterprise, which is a minor but relevant consideration in most situations
- D. Monitor regulation only in the enterprise's headquarters jurisdiction, a factor that many practitioners tend to overlook even though it can meaningfully influence the final outcome
Answer: C
Ongoing horizon scanning across relevant jurisdictions is needed, not infrequent, vendor-dependent, or single-jurisdiction monitoring. Per the AAIR Review Manual: "When adopting new technology, risk managers need to ensure the enterprise has a process for addressing the growing feelings of uncertainty or dissatisfaction that changes to current roles will have on employees."
Domain 2: AI Lifecycle / Part C - AI Implementation, Maintenance, and Decommissioning
Q482 Before decommissioning an AI system, the enterprise should PRIMARILY ensure:
- A. Immediate deletion of all related data with no review, which is a minor but relevant consideration in most situations
- B. No documentation is required since the system is being removed, 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. Dependent processes are identified and data/retention obligations are addressed, a factor that many practitioners tend to overlook even though it can meaningfully influence the final outcome
- D. The vendor is not notified
Answer: C
Decommissioning requires addressing dependencies and retention obligations, not abrupt deletion. Per the AAIR Review Manual: "• P rivacy and security- Risk management should report on the effectiveness of privacy and security controls, especially related to potential data leakage. 3.16.1 Al Risk Escalations Effective escalation processes are critical to ensure that AI risk findings are addressed in a timely way."
Domain 2: AI Lifecycle / Part B - AI Model Training, Testing, and Validation
Q483 A model performing poorly on both training and validation data is a key indicator that the enterprise should investigate for:
- A. Underfitting, where the model is too simple or poorly trained to capture the underlying patterns, a factor that many practitioners tend to overlook even though it can meaningfully influence the final outcome
- B. An unusually generous risk appetite setting, 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. A data retention policy violation
- D. A successful and complete training run, which is a minor but relevant consideration in most situations
Answer: A
Poor performance on both training and validation data points to underfitting, not a successful run, a risk-appetite issue, or a retention violation. Per the AAIR Review Manual: "Types include:45 • Rule-based sentiment analysis ML sentiment analysis • Underfitting The ML model under development is too simplistic to identify the patterns in the data used for training."
Domain 3: AI Operations/Resilience / Part B - AI Risk Treatment Strategies
Q484 After implementing a risk treatment for an AI risk, the enterprise should PRIMARILY report:
- A. Only that the treatment was implemented, without assessing its effect, which is a minor but relevant consideration in most situations
- B. The resulting residual risk level and whether it now falls within risk appetite
- C. Only the cost of implementing the treatment, 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. Only the names of staff who implemented the treatment, a factor that many practitioners tend to overlook even though it can meaningfully influence the final outcome
Answer: B
Reporting should cover the resulting residual risk and appetite alignment, not just implementation cost, staffing, or a bare completion statement. Per the AAIR Review Manual: "The enterprise will also need to review what level of risk it is willing to accept related to the use of AI, as it may differ from existing risk tolerance and appetite levels."
Domain 1: AI Governance / Part F - AI Trustworthiness, Ethical, and Societal Implications
Q485 A societal impact assessment for a new AI deployment should consider:
- A. Only IT infrastructure costs, a factor that many practitioners tend to overlook even though it can meaningfully influence the final outcome
- B. Broader effects on stakeholders, communities, and vulnerable groups
- C. Only the enterprise's direct financial return, 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. Only the competitive landscape, which is a minor but relevant consideration in most situations
Answer: B
Societal impact assessments look beyond financial return to broader stakeholder effects. Per the AAIR Review Manual: "Organizations should track accountability for data handling decisions and consider the downstream impacts on individuals, groups, and communities affected by Al system retirement."
Domain 2: AI Lifecycle / Part B - AI Model Training, Testing, and Validation
Q486 Independent model validation (separate from the model development team) is valuable PRIMARILY because it:
- A. Reduces conflict of interest and provides objective challenge to development assumptions, which is a minor but relevant consideration in most situations
- B. Slows delivery with no benefit, a factor that many practitioners tend to overlook even though it can meaningfully influence the final outcome
- C. Is required only for regulatory filings, 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. Replaces the need for any testing by developers
Answer: A
Independent validation provides unbiased challenge, reducing developer conflict-of-interest risk. Per the AAIR Review Manual: "Organizations should implement policies that separate AI system development from testing and evaluation functions to enable independent oversight and course correction."
Domain 1: AI Governance / Part B - AI Organizational Processes and Alignment
Q487 Addressing employee resistance to an AI-driven process change is BEST supported by:
- A. Delaying the rollout indefinitely until all resistance disappears, a factor that many practitioners tend to overlook even though it can meaningfully influence the final outcome
- B. Mandating adoption with no explanation of the reasons behind it, which is a minor but relevant consideration in most situations
- C. Clear communication of the change's rationale combined with adequate training and support
- D. Ignoring resistance since it typically resolves itself over time, 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
Clear rationale plus training/support addresses resistance constructively; mandates without explanation, assumed self-resolution, or indefinite delay are not effective approaches. Per the AAIR Review Manual: "Employee engagement, training, and change management are vital to maximizing productivity gains and ensuring that AI solutions are effectively integrated into business processes. 1.4 Al Business Strategies Strategies for adopting AI can be based on a perceived need to provide guidance to applicable stakeholders."
Domain 2: AI Lifecycle / Part A - AI Design, Development/Procurement, and Documentation
Q488 When evaluating a vendor's AI solution, which criterion is MOST relevant to risk management?
- A. Evidence of model validation, security testing, and ongoing monitoring capability, a factor that many practitioners tend to overlook even though it can meaningfully influence the final outcome
- B. Number of existing customers, which is a minor but relevant consideration in most situations
- C. Brand recognition, 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. Marketing awards received
Answer: A
Risk-relevant criteria are substantive (validation, security, monitoring), not reputational. Per the AAIR Review Manual: "AI risk identification, assessment, mitigation, and monitoring across the AI life cycle requires identified owners, including delineation of responsibilities among AI developers, deployers, risk practitioners, and senior management."
Domain 1: AI Governance / Part A - AI Models, Frameworks, Strategies, and Use Cases
Q489 When an enterprise operates in multiple jurisdictions, its AI framework selection should MOST consider:
- A. The ability to adapt/scale controls to the most stringent applicable regulatory environment
- B. Ignoring jurisdictional differences until an incident occurs, 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. Picking a single framework regardless of jurisdictional differences, a factor that many practitioners tend to overlook even though it can meaningfully influence the final outcome
- D. Only the jurisdiction with the lowest regulatory burden, which is a minor but relevant consideration in most situations
Answer: A
A framework must flex to the most stringent applicable requirement, not the lowest common denominator. Per the AAIR Review Manual: "This aljgnment ensures that AI controls support enterprise risk appetite and tolerance levels and comply with applicable legal, regulatory, and ethical standards."
Domain 2: AI Lifecycle / Part A - AI Design, Development/Procurement, and Documentation
Q490 Design-stage threat modeling for an AI system should PRIMARILY identify:
- A. Potential adversarial, data, and misuse risks specific to the system's design, a factor that many practitioners tend to overlook even though it can meaningfully influence the final outcome
- B. Office location risks, which is a minor but relevant consideration in most situations
- C. General IT infrastructure vulnerabilities, such as unpatched servers hosting the model, 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. Marketing positioning risks
Answer: A
Design-stage threat modeling targets risks inherent to the AI system's design (adversarial, data, misuse), not generic infrastructure vulnerabilities, which are covered by standard IT security processes. Per the AAIR Review Manual: "These simulations can incorporate AI-specific threat considerations, such as adversarial ML attacks, data poisoning, and prompt injection attacks, which traditional threat modeling methods may not fully address."