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 2: AI Lifecycle / Part D - AI Data and Asset Management
Q311 Data quality controls for AI inputs should PRIMARILY address:
- A. Accuracy, completeness, and representativeness of the data used, which is a minor but relevant consideration in most situations
- B. The visual formatting of data files
- C. The volume of data collected, regardless of relevance, 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. The color-coding of data dashboards, a factor that many practitioners tend to overlook even though it can meaningfully influence the final outcome
Answer: A
Data quality controls target substantive attributes (accuracy, completeness, representativeness) that affect model outcomes; data volume alone, without regard to relevance, does not ensure quality. Per the AAIR Review Manual: "AI developers and users need to ensure they fully understand the data flow and integrity controls from the source systems being input into the model, as well as data completeness, accuracy, and cleansing methods for using data to train and then put through the models."
Domain 2: AI Lifecycle / Part C - AI Implementation, Maintenance, and Decommissioning
Q312 Deprecation notices to end users before retiring an AI-powered feature are PRIMARILY important because they:
- A. Give users time to adapt workflows and reduce disruption from the feature's discontinuation, which is a minor but relevant consideration in most situations
- B. Are only required for consumer-facing, not internal, AI tools, 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. Replace the need for a technical decommissioning plan
- D. Are optional if the feature was rarely used, a factor that many practitioners tend to overlook even though it can meaningfully influence the final outcome
Answer: A
Advance deprecation notices allow affected users time to adjust, reducing operational disruption; they are a user-communication step alongside, not a substitute for, the technical decommissioning plan.
Domain 3: AI Operations/Resilience / Part A - AI Risk Scenario Identification and Assessment
Q313 Developing AI risk scenarios (e.g., biased lending decisions) is MOST useful for:
- A. Replacing the need for controls, which is a minor but relevant consideration in most situations
- B. Satisfying only external auditors, a factor that many practitioners tend to overlook even though it can meaningfully influence the final outcome
- C. Making the risk register appear larger
- D. Translating abstract AI risk into concrete, assessable situations for the enterprise, 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
Concrete scenarios make abstract AI risk assessable and actionable. Per the AAIR Review Manual: "Increasing dataset quantity and diversity also helps reduce bias in decisions, especially if certain segments are underrepresented in the original data, and can lead to better and more accurate decisions by providing a wider range of real-world situations."
Domain 3: AI Operations/Resilience / Part E - AI Supply Chain Risk Management
Q314 Regarding AI vendor security assessment, the BEST practice is to:
- A. Conduct security assessments of AI vendors covering data handling, model security, and incident history, a factor that many practitioners tend to overlook even though it can meaningfully influence the final outcome
- B. Rely solely on a vendor's self-certification of security practices
- C. Limit vendor security assessment to contractual penalty clauses, 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. Assess vendor security only once, at initial onboarding, which is a minor but relevant consideration in most situations
Answer: A
Vendor security assessment should be substantive and periodically refreshed, not based solely on self-certification or onboarding-only. Per the AAIR Review Manual: "A thorough control gap assessment begins with an accurate understanding of the current control environment, which can be established through various sources such as audits, control tests, incident reports, vulnerability assessments, and direct observations."
Domain 1: AI Governance / Part B - AI Organizational Processes and Alignment
Q315 Communicating planned AI changes to affected staff before rollout is valuable PRIMARILY because it:
- A. Is unnecessary if the change is technically minor, 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. Reduces resistance and surprises by setting expectations ahead of time
- C. Guarantees staff will support the change, which is a minor but relevant consideration in most situations
- D. Replaces the need for user training, a factor that many practitioners tend to overlook even though it can meaningfully influence the final outcome
Answer: B
Advance communication manages expectations and resistance; it doesn't guarantee support or replace training.
Domain 3: AI Operations/Resilience / Part A - AI Risk Scenario Identification and Assessment
Q316 Including prompt injection as a documented risk scenario for generative AI deployments is important PRIMARILY because it:
- A. Guarantees prompt injection attacks will never succeed
- B. Ensures the enterprise has considered and planned mitigations for this specific attack vector in advance, 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. Removes the need for output content filtering, a factor that many practitioners tend to overlook even though it can meaningfully influence the final outcome
- D. Is only relevant to publicly accessible chatbots, which is a minor but relevant consideration in most situations
Answer: B
Documenting the scenario ensures proactive mitigation planning; it doesn't guarantee prevention, isn't limited to public chatbots, or replace output filtering. Per the AAIR Review Manual: "Attack Name Attack Description Prompt injection A prompt injection occurs when an attacker uses specific prompts to alter the behavior of a large language model (LLM), often nefariously."
Domain 3: AI Operations/Resilience / Part E - AI Supply Chain Risk Management
Q317 Regarding AI supply chain risk mapping, the BEST practice is to:
- A. Map the full AI supply chain, including sub-vendors and third-party data/model providers, not just direct vendors, a factor that many practitioners tend to overlook even though it can meaningfully influence the final outcome
- B. Map the supply chain only once at initial vendor onboarding
- C. Assess risk only for the primary contracted vendor, which is a minor but relevant consideration in most situations
- D. Assume sub-vendor risk is immaterial if not contractually visible, 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: A
Full supply chain mapping, including sub-vendors, is needed and should be kept current, not limited to direct vendors or a one-time exercise. Per the AAIR Review Manual: "Given the layered nature of AI supply chains, where third-party and fourth-party vendors may also incorporate open-source software, the risk of supply chain compromise is amplified."
Domain 3: AI Operations/Resilience / Part C - AI Controls Management
Q318 When addressing control framework mapping for AI, what is the BEST first step for an AI risk practitioner?
- A. Map controls to a framework only for externally audited systems, 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. Design AI controls ad hoc without reference to any framework
- C. Treat framework mapping as a one-time exercise never revisited, a factor that many practitioners tend to overlook even though it can meaningfully influence the final outcome
- D. Map AI controls to a recognized control framework to ensure completeness and avoid gaps, which is a minor but relevant consideration in most situations
Answer: D
Mapping to a recognized framework, kept current, supports completeness; ad hoc or one-time-only mapping risks gaps. Per the AAIR Review Manual: "3.11 .1 Control Gap Analysis An important first step when determining what controls are needed for AI solutions is to identify any gaps in existing controls that are created by the risk presented by the use of AI."
Domain 2: AI Lifecycle / Part A - AI Design, Development/Procurement, and Documentation
Q319 A model design document lacking documented assumptions and limitations MOST increases the risk of:
- A. Improved explainability, which is a minor but relevant consideration in most situations
- B. Users applying the model outside its valid intended scope, a factor that many practitioners tend to overlook even though it can meaningfully influence the final outcome
- C. Reduced training cost
- D. Faster deployment with no downside, 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
Undocumented limitations increase risk of out-of-scope or inappropriate model use. Per the AAIR Review Manual: "Key components of AI model documentation include: 102 • The model used • The data used to develop it • Data preprocessing and cleansing steps taken • Testing and experiments performed • Intended use cases and assumptions • Performance metrics • Limitations • Stakeholders • Version history • Fairness and bias testing"
Domain 1: AI Governance / Part C - AI Ownership, Oversight, and Accountability
Q320 When an AI system causes harm to a customer, accountability MOST appropriately resides with:
- A. No one, since AI decisions are autonomous, 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. The data science team that built the model, since they best understand its technical limitations
- C. The accountable business/model owner who approved its use, regardless of technical cause, a factor that many practitioners tend to overlook even though it can meaningfully influence the final outcome
- D. The AI model itself, which is a minor but relevant consideration in most situations
Answer: C
Accountability rests with the owner who approved the system's use, not with the builders (technical knowledge isn't the basis for accountability), the model itself, or no one. Per the AAIR Review Manual: "Unlike traditional human actions where it is clear who is responsible, AI systems can act autonomously, raising critical questions about who is responsible and accountable when AI causes harm or violates laws."