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 1: AI Governance / Part C - AI Ownership, Oversight, and Accountability
Q211 An executive sponsor for an AI initiative should PRIMARILY be accountable for:
- A. Writing the model's technical specification, which is a minor but relevant consideration in most situations
- B. Approving the model's source code directly
- C. Performing the model's statistical validation personally, a factor that many practitioners tend to overlook even though it can meaningfully influence the final outcome
- D. Ensuring the initiative's risk is managed in line with its business value and the enterprise's risk appetite, 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
Executive sponsors own risk-vs-value alignment at the governance level, not hands-on technical validation tasks. Per the AAIR Review Manual: "This makes the review of new and existing vendor contracts key to ensuring legal and regulatory concepts are adequately addressed in line with the enterprise's risk tolerance and appetite."
Domain 3: AI Operations/Resilience / Part A - AI Risk Scenario Identification and Assessment
Q212 A model poisoning attack is a risk the enterprise should guard against PRIMARILY because an attacker who tampers with training could:
- A. Make the model easier for others to audit, which is a minor but relevant consideration in most situations
- B. Increase the enterprise's cloud hosting bill
- C. Cause the model to behave incorrectly or maliciously under attacker-chosen conditions, 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. Reduce the model's file size on disk, a factor that many practitioners tend to overlook even though it can meaningfully influence the final outcome
Answer: C
Poisoning lets an attacker engineer malicious or incorrect behavior under chosen conditions, not file size, cost, or auditability effects. Per the AAIR Review Manual: "Intentional changes, such as model poisoning, involve malicious actors manipulating the training process to alter model behavior."
Domain 1: AI Governance / Part F - AI Trustworthiness, Ethical, and Societal Implications
Q213 When an AI system's output could plausibly cause reputational or societal harm even if technically accurate, the organization's ethical governance process should PRIMARILY:
- A. Evaluate the broader societal and reputational impact alongside technical accuracy before deployment, 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. Defer the decision entirely to the legal department with no governance input, a factor that many practitioners tend to overlook even though it can meaningfully influence the final outcome
- C. Proceed with deployment since technical accuracy is the only relevant standard
- D. Only consider societal impact after the system has already caused harm, which is a minor but relevant consideration in most situations
Answer: A
Ethical governance requires evaluating broader societal and reputational consequences in addition to technical accuracy, rather than treating accuracy as sufficient on its own. Per the AAIR Review Manual: "This should be done through the organization's data governance process and could include standard techniques, such as performing a data protection impact assessment (DPIA )."
Domain 2: AI Lifecycle / Part B - AI Model Training, Testing, and Validation
Q214 When addressing independent validation of model results, what is the BEST first step for an AI risk practitioner?
- A. Have model validation performed by a function independent from model development, which is a minor but relevant consideration in most situations
- B. Allow the development team to self-validate without independent review, 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. Skip independent validation for internally facing tools, a factor that many practitioners tend to overlook even though it can meaningfully influence the final outcome
- D. Use independent validation only for regulator-mandated models
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 F - AI Trustworthiness, Ethical, and Societal Implications
Q215 What is the most effective way to manage AI fairness across demographic groups within an AI risk program?
- A. Test AI outcomes for disparate impact across relevant demographic groups before and after deployment, a factor that many practitioners tend to overlook even though it can meaningfully influence the final outcome
- B. Assume fairness if overall accuracy is high, which is a minor but relevant consideration in most situations
- C. Limit fairness testing to the development dataset only, 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. Test for fairness only if a complaint is received
Answer: A
Fairness testing should proactively assess subgroup outcomes, not rely on aggregate accuracy or reactive complaints.
Domain 3: AI Operations/Resilience / Part C - AI Controls Management
Q216 Regularly rotating API keys used to access AI services is valuable PRIMARILY because it:
- A. Automatically improves the AI service's response quality
- B. Removes the need for monitoring API usage, a factor that many practitioners tend to overlook even though it can meaningfully influence the final outcome
- C. Is only necessary if the enterprise has experienced a prior breach, which is a minor but relevant consideration in most situations
- D. Limits the window of exposure if a key is ever compromised or leaked, 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
Rotation limits exposure if a key leaks, not response quality; it's good practice regardless of prior breach history and doesn't replace usage monitoring.
Domain 2: AI Lifecycle / Part A - AI Design, Development/Procurement, and Documentation
Q217 Documenting training data provenance is MOST important for:
- A. Enabling faster model training through better-organized data, 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. Understanding data quality, bias, and legal rights to use the data, a factor that many practitioners tend to overlook even though it can meaningfully influence the final outcome
- C. Satisfying only internal curiosity
- D. Reducing storage requirements, which is a minor but relevant consideration in most situations
Answer: B
Provenance documentation underpins data quality, bias, and legal-use assessments; faster training is a data-organization benefit, not provenance's primary purpose. Per the AAIR Review Manual: "• Bias detection and fairness-Training programs should provide employees with an understanding of how bias in data and model outputs can lead to unfair or harmful outcomes."
Domain 3: AI Operations/Resilience / Part E - AI Supply Chain Risk Management
Q218 Exit/transition planning for a critical AI vendor relationship should be established:
- A. Proactively, as part of onboarding, to avoid disruption if a transition becomes necessary
- B. Only after the vendor relationship has already failed, a factor that many practitioners tend to overlook even though it can meaningfully influence the final outcome
- C. Never, since switching vendors is not realistic, which is a minor but relevant consideration in most situations
- D. Only if required by the vendor's contract template, 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
Proactive exit planning reduces disruption risk if a transition later becomes necessary.
Domain 2: AI Lifecycle / Part D - AI Data and Asset Management
Q219 Defining a retention policy for user prompts submitted to a generative AI application is valuable PRIMARILY because it:
- A. Limits how long potentially sensitive user input is stored, reducing exposure risk
- B. Automatically improves the relevance of the model's responses, a factor that many practitioners tend to overlook even though it can meaningfully influence the final outcome
- C. Removes the need for access controls on stored logs, which is a minor but relevant consideration in most situations
- D. Is only relevant for applications used by minors, 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
A retention policy limits storage duration of potentially sensitive input, reducing exposure; it doesn't affect response relevance, isn't minor-specific, or remove the need for access controls. Per the AAIR Review Manual: "This approach mitigates privacy risk by reducing the volume of sensitive or personal information exposed to potential breaches or misuse."
Domain 1: AI Governance / Part B - AI Organizational Processes and Alignment
Q220 Assessing whether the AI governance program is adequately resourced should PRIMARILY consider:
- A. Whether every other enterprise in the industry has a similar program
- B. Whether staffing and budget are sufficient relative to the scale and risk of the enterprise's AI use, 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. Whether the program's budget is lower than that of the data science team, a factor that many practitioners tend to overlook even though it can meaningfully influence the final outcome
- D. Whether the program has existed for at least two full years, which is a minor but relevant consideration in most situations
Answer: B
Adequacy should be judged against the enterprise's actual scale/risk of AI use, not relative budget comparisons, program age, or industry conformity alone. Per the AAIR Review Manual: "Overview An essential part of managing artificial intelligence (AI) risk is understanding how the use of AI can be integrated into an enterprise's existing governance and risk management program."