About these questions
These five questions are original. We wrote them to match the style of the AAIR content outline, with the domains roughly in proportion to their weights (two from Domain 1, one from Domain 2, two from Domain 3). They are not real exam questions, and the real exam will word things differently. Every question has one best answer among four options, which is the format ISACA describes in its candidate guide, so there are no choose-two questions here.
How to use these
Try each question without your notes, and choose an answer before you open the explanation. Then read the explanation even when you were right, because the reason the closest wrong option fails is where the learning is. Note any domain where you slipped and revisit it on the domains explained page. Watch for words such as BEST and FIRST. They decide between answers that are all partly true.
Question 1 (AI Risk Governance and Framework Integration)
The executive team states that the company will accept only a low level of risk from AI used in hiring decisions. The AI risk lead must turn this statement into something that the owner of the screening model can monitor each month. What should the lead produce?
- A. A risk capacity figure based on the company’s financial reserves
- B. A taxonomy listing every AI control used in recruitment
- C. A broader appetite statement covering all AI use in the company
- D. Risk tolerance levels with measurable thresholds for the model’s key metrics
Show the answer and explanation
Correct answer: D. Risk appetite is the high-level statement of how much risk the organization is willing to take. Risk tolerance turns it into acceptable deviations per objective or process, with numbers that a model owner can track and escalate against. Risk capacity, the closest wrong option, is the most risk the organization could absorb before its survival is threatened. It is a ceiling, not a monitoring threshold, and it says nothing about how much risk management chooses to accept. A control taxonomy classifies controls but sets no limits.
Question 2 (AI Risk Governance and Framework Integration)
An online lender plans to introduce an AI system that estimates whether private individuals are creditworthy and sets their credit limits. The lender operates in the European Union. Under the EU AI Act, which description fits this system best?
- A. A high-risk system, because it affects access to essential private services
- B. A limited-risk system, because it only needs a transparency notice
- C. A minimal-risk system, because the decision is made by software
- D. A prohibited system, because it scores people automatically
Show the answer and explanation
Correct answer: A. Annex III of the EU AI Act lists AI systems that evaluate the creditworthiness of natural persons or establish their credit score under the area of access to essential private and public services. Such systems are high-risk, which brings duties for the provider and the deployer, including risk management, documentation, and human oversight. Limited risk is the closest wrong option: it describes systems with transparency duties such as chatbots, which is a separate category. Credit scoring is not on the prohibited list, so option D is wrong. Note that the Act has exceptions, such as systems used only to detect financial fraud, which this scenario does not describe.
Question 3 (AI Life Cycle Risk Management)
A school district trains a model to find students who need extra tutoring. The training label is the number of times each student has visited the tutoring center in the past. Reviewers find that students with long bus rides rarely visit, so the model keeps missing them even though they struggle academically. Which problem does this describe?
- A. Overfitting, because the model learned noise from its training records
- B. Data leakage, because evaluation data reached the training process
- C. Measurement bias, because the recorded proxy systematically differs from the real need
- D. Concept drift, because the relationship between inputs and outcomes changed after deployment
Show the answer and explanation
Correct answer: C. Measurement bias appears when the quantity that is recorded stands in for the thing you actually care about, and the two diverge in a systematic way. Visits to the center reflect access and transport as well as need, so the label builds those patterns into the model. Concept drift is the closest wrong option. It means that relationships change over time after deployment, but this model was flawed from the start, and nothing in the scenario changed afterward. Overfitting and data leakage are about how the model was fitted and tested, not about what the label represents.
Question 4 (AI Risk Program Management)
An insurer has budget for one control on an AI agent that approves and pays out small claims. Design one blocks a payment above a set amount until a person approves it. Design two sends the claims team a report of unusual payments the next morning. Money that has been paid out cannot be recovered. Which choice is BEST?
- A. Design two, because a daily report also creates an audit trail
- B. Design one, because the loss is irreversible once a payment is made
- C. Design two, because it costs less to run than a manual approval
- D. Neither, because a single control cannot cover the exposure
Show the answer and explanation
Correct answer: B. A preventive control stops the event, while a detective control only reports it afterward. When the loss cannot be undone and there is budget for just one control, prevention removes exposure that detection could only record. Design two, the closest wrong option, would give you good evidence of what went wrong but no way to get the money back. Cost and audit trail are real benefits, but they do not outweigh the constraint in the stem. Option D ignores that you must decide with the budget you have. Detection earns priority when prevention is not possible or would be too disruptive.
Question 5 (AI Risk Program Management)
A risk committee wants one measure to show whether the required sign-off before each model release is actually being completed. Which measure works best as a key control indicator?
- A. The percentage of model releases with a recorded sign-off before go-live
- B. The number of AI incidents logged by customers this quarter
- C. The number of models currently listed in the AI inventory
- D. The change in average user satisfaction scores after each release
Show the answer and explanation
Correct answer: A. A key control indicator tells you whether a specific control is operating as intended. The share of releases with a sign-off on record answers that question directly, and release records can serve as evidence. Incidents reported by customers are the closest wrong option: they work as a risk or outcome indicator that may point to a problem, but they cannot show whether the sign-off took place. The inventory count measures a different control, and satisfaction scores measure performance, not control operation.
More guides for AAIR
- AAIR Study Plan: 4-Week and 8-Week Schedules
- Is the AAIR Exam Hard? An Honest Look
- AAIR Prerequisites and Eligibility Explained
- AAIR Renewal and Retakes Explained
- Is AAIR Worth It? An Honest Look
- AAIR Cheat Sheet: Key Facts to Review
- AAIR vs AAISM: Which ISACA AI Exam First?
- AAIA vs AAIR: Which ISACA AI Exam First?
Next step
Read the exam format page if you want to know how the real exam presents questions. For more practice in this style, with explanations on every option, try the AAIR practice exams.