Free AAIA Quick Start: 10 Mixed Practice Questions

This diagnostic includes AI Governance and Risk, AI Operations, and AI Auditing Tools and Techniques. FlashGenius readiness results are guidance, not official ISACA passing scores.

Review all AAIA domains or begin the interactive quick-start test.

Quick-start questions with answers

Sample Question 1 — AI Operations

A telecommunications company uses an AI churn model to drive retention offers. The model performs well overall, but several upstream source fields were recently changed. Data teams provide an outdated lineage diagram and state that quality checks ran successfully. The auditor wants to conclude whether production scores issued last month can be traced to the source data used. Which evidence BEST supports the conclusion?

  1. A. Run-level metadata and reconciliations linking scores to source snapshots. (Correct answer)
  2. B. Approved lineage diagrams showing intended source-to-feature data flows.
  3. C. Access review records for users with feature table modification rights.
  4. D. Quality dashboard results showing successful checks for the scoring period.

Correct answer: A

Explanation: A is correct because it provides direct evidence of actual historical traceability from specific production scores back to the source snapshots used. That is the strongest support for reconstructing what data drove decisions. B is wrong because lineage diagrams describe intended flows rather than proving what data was actually used in a given run. C is wrong because access reviews help assess unauthorized-change risk but do not identify the exact input version tied to scored outputs. D is wrong because successful quality checks show checks occurred, not that individual scores can be traced to the precise source data used.

Sample Question 2 — AI Auditing Tools and Techniques

A bank uses a third-party explainability dashboard for an AI credit decision model. Management asserts the dashboard supports transparency controls for adverse action reviews. The dashboard provides monthly global feature importance, while case-level explanations are retained only for selected decisions. The vendor documentation warns that global importance should not be used to explain individual outcomes. What is the PRIMARY audit concern?

  1. A. Vendor limitations were not incorporated into the transparency control assessment
  2. B. Global outputs are relied on for decision-level transparency evidence (Correct answer)
  3. C. Case-level explanation records are retained for only selected decisions
  4. D. Manual adverse action narratives are used when explanations are unavailable

Correct answer: B

Explanation: B is best because the core design issue is that the evidence being relied on does not fit the control objective. A control intended to support transparency for individual credit decisions cannot be supported primarily by global feature importance, especially when vendor documentation states that such output should not be used to explain single outcomes. A is a related governance weakness, C is an important evidence-retention gap, and D may indicate compensating manual effort, but each is secondary to the fundamental mismatch between global explainability output and decision-level transparency requirements.

Sample Question 3 — AI Governance and Risk

A health insurer piloting AI-assisted claims triage waived several required predeployment controls to meet a launch deadline. The exception register exists, but several entries have no expiry date and were approved by the business sponsor rather than the required AI risk committee. What is the PRIMARY audit concern?

  1. A. Exceptions lack required authority and defined expiration for high-risk use cases. (Correct answer)
  2. B. Efficiency benefits were considered when management assessed residual risk levels.
  3. C. Exception reporting groups multiple control waivers into summary categories.
  4. D. Compensating controls rely on narrative descriptions from control owners.

Correct answer: A

Explanation: A is correct because the most significant failure is that risk exceptions for a high-risk AI use case were accepted outside required authority and without clear time limits. That indicates the exception process is not operating as intended. B may be a concern, but it is secondary to unauthorized and open-ended risk acceptance. C can reduce transparency, but reporting aggregation is less severe than invalid approvals. D reflects weak supporting evidence, but the larger issue is that the exceptions themselves were not properly authorized or bounded.

Sample Question 4 — AI Operations

During an audit of a healthcare diagnostic AI model, the auditor finds that model performance monitoring focuses on overall accuracy and latency, but there is no monitoring for data drift or concept drift. The model was trained on data from two years ago, and the hospital has recently changed diagnostic protocols. Which of the following is the MOST appropriate audit objective in this situation?

  1. A. To determine whether the AI model complies with GDPR requirements for data minimization and purpose limitation.
  2. B. To assess whether the AI model’s lifecycle management includes controls to detect and respond to data and concept drift that could impact clinical decision quality. (Correct answer)
  3. C. To verify that the AI model’s training data was obtained with appropriate patient consent and ethical approvals.
  4. D. To evaluate whether the AI model’s source code is protected against unauthorized modification in line with ISO 27001.

Correct answer: B

Explanation: The scenario highlights a change in clinical protocols and the absence of drift monitoring, which directly affects the reliability and safety of AI outputs in a healthcare context. In AI Operations, drift monitoring is critical to ensure models remain valid as environments and practices evolve. Option B is correct because it directly targets lifecycle management controls for detecting and responding to data and concept drift, which is the key operational risk in this scenario. This aligns with NIST AI RMF and ISO/IEC 42001 requirements for ongoing monitoring and performance management, especially for high-impact use cases like healthcare. Option A (GDPR data minimization and purpose limitation) is important but not the primary concern raised by the facts given. The issue is not about lawful basis or purpose, but about model degradation due to environmental changes. Option C (consent and ethics) is relevant to governance and compliance but does not address the operational risk that the model may now be misaligned with current clinical practice. Option D (source code protection) relates to security and integrity but does not address the risk that the model’s predictions may be clinically inappropriate due to drift. The PRIMARY audit objective should focus on lifecycle and monitoring controls for drift in this context.

Sample Question 5 — AI Auditing Tools and Techniques

A global bank retrains its AML anomaly-detection model quarterly. A lineage platform maps automated training data feeds, and retraining approvals reference the lineage report. However, analyst-uploaded reference files are sometimes used during retraining and are not captured by the platform. Model performance has remained stable, and management states the uploads are rare. Which audit procedure is MOST appropriate?

  1. A. Reconcile the lineage report to approved source inventory and manual upload logs for recent retraining (Correct answer)
  2. B. Compare post-retraining performance metrics with prior quarters to identify unexplained changes
  3. C. Inspect retraining approval packets to confirm management signed off before production release
  4. D. Interview analysts about how often offline reference files are used during retraining

Correct answer: A

Explanation: A is best because it directly tests whether lineage coverage is complete across both automated and manual inputs by reconciling the lineage output to approved sources and upload records. B evaluates outcomes, but stable performance does not show that all training data was approved and traceable. C checks approval mechanics, yet those approvals may rely on an incomplete lineage report. D can provide useful context, but inquiry is weaker evidence than reconciliation to traceable records.

Sample Question 6 — AI Governance and Risk

A consumer lender uses AI to prioritize loan applications for manual underwriting review. Underwriters must click final approval, but review logs show near-total acceptance of AI recommendations. Reviewers process high volumes under strict service levels, and training focuses mainly on tool navigation. Which control deficiency is MOST significant?

  1. A. Reviewer workload and training limit meaningful challenge of AI recommendations. (Correct answer)
  2. B. Low complaint volumes are not reconciled to underwriting decision records.
  3. C. Override rates are reported without comparison to historical approval patterns.
  4. D. Final approval clicks are captured without separate productivity reporting.

Correct answer: A

Explanation: A is correct because the core issue is whether the supposed human review can function as a real control. Near-total acceptance, high throughput pressure, and minimal challenge-oriented training indicate the human step may be only nominal. B is a monitoring weakness, but not the main deficiency in the review control itself. C could improve analysis, yet it is secondary to whether reviewers have the capacity and competence to challenge outputs at all. D is administrative and does not address the effectiveness of human oversight.

Sample Question 7 — AI Auditing Tools and Techniques

A manufacturing company uses an automated bot to collect monthly configuration evidence from production systems for AI infrastructure control testing. The auditor wants to determine whether the bot operated effectively during the audit period. Which audit procedure is MOST appropriate?

  1. A. Interview the control owner about the bot's normal monthly process.
  2. B. Review the bot design document and approved collection schedule.
  3. C. Confirm the bot account has read-only access to production systems.
  4. D. Reconcile bot run logs to source records for a sample of months. (Correct answer)

Correct answer: D

Explanation: D is best because testing operating effectiveness requires evidence that the bot actually ran as intended during the period and collected reliable output. Reconciling run logs to source records for selected months provides direct evidence of execution, timing, and collection accuracy. A and B help the auditor understand the process and design, but they do not prove operation. C addresses permissions, which is relevant, but it does not show the bot executed completely and accurately.

Sample Question 8 — AI Governance and Risk

A payment processor deployed a major update to an AI fraud scoring model before a peak sales period. Production logs show deployment occurred five days before independent validation was completed. Fraud losses declined after release, and an emergency change ticket was approved by the same executive sponsor accountable for fraud reduction targets. Which control deficiency is MOST significant?

  1. A. The model update entered production before required independent validation and approval were completed. (Correct answer)
  2. B. The emergency change criteria were ambiguous and allowed inconsistent business interpretation.
  3. C. The executive sponsor had approval authority while also owning fraud reduction objectives.
  4. D. The post-deployment results emphasized fraud losses without equal focus on other outcomes.

Correct answer: A

Explanation: A is the best answer because the key preventive governance control failed: a material model change reached production before completion of the required independent challenge process. In a high-impact AI use case, favorable short-term results do not offset a bypass of pre-deployment validation and approval gates. B is a meaningful design weakness, but it is secondary to the confirmed failure of the actual release control. C is also a valid segregation-of-duties concern, but it is less significant than the fact that production deployment occurred before required independent review was complete. D identifies a monitoring or reporting weakness, not the primary control failure that allowed the model into production.

Sample Question 9 — AI Operations

A healthcare organization monitors a diagnostic support model for data drift. Drift alerts are generated weekly, but several alerts remain open for more than 60 days because the operations team and clinical model owner disagree on who must decide whether retraining is needed. Which recommendation is MOST appropriate?

  1. A. Define accountable alert owners with escalation and disposition timeframes (Correct answer)
  2. B. Increase the statistical sensitivity of drift thresholds for clinical variables
  3. C. Require quarterly committee review of all model monitoring dashboards
  4. D. Add additional model performance metrics to the production dashboard

Correct answer: A

Explanation: A is the best recommendation because the failure is not detection; alerts already exist. The operating-effectiveness gap is unclear ownership and no timely disposition process, so accountability, escalation, and closure timeframes are needed. B targets threshold design, but the scenario shows the control is already detecting issues. C adds governance oversight, but quarterly review is too indirect to fix unresolved operational alerts. D may expand monitoring, but more metrics do not resolve the existing failure to assign responsibility and act on generated alerts.

Sample Question 10 — AI Governance and Risk

A retail bank deploys a vendor-hosted generative AI assistant to support customer service agents. Procurement completed a standard vendor security review, and the service has met availability targets for six months. The contract allows the vendor to change the underlying model and service behavior without prior notice, and it does not specify AI-related audit rights or data-use restrictions. What is the PRIMARY audit concern?

  1. A. The organization may lack enforceable oversight of vendor changes, data use, and AI-specific assurance obligations. (Correct answer)
  2. B. The organization may be relying on availability results instead of validating the assistant's response accuracy.
  3. C. The organization may not have sufficient internal expertise to independently review the vendor's model design.
  4. D. The organization may be using a standard procurement workflow instead of a technology-specific onboarding path.

Correct answer: A

Explanation: A is best because the main governance risk is the absence of enforceable third-party AI obligations. A standard security review and good availability performance do not address AI-specific risks such as provider model changes, data usage, audit rights, notification duties, and assurance access. The subtle distinction is that the deficiency is not simply poor vendor performance; it is weak control design over outsourced AI risk. Why the other options are weaker: B) Response quality matters, but the scenario's strongest concern is the inability to govern vendor-controlled changes and obligations. C) Lack of internal model-design expertise may be relevant for some AI audits, but it is not the primary issue when contractual governance rights are absent. D) The audit issue is not the workflow label; it is the lack of AI-specific contractual controls and oversight rights.

AAIA Practice Test FAQs

What is the ISACA Advanced in AI Audit (AAIA) exam?

AAIA is an ISACA certification exam focused on auditing artificial intelligence systems. It covers AI governance and risk, AI operations, and AI auditing tools and techniques.

How many questions and how much time does the AAIA exam have?

The AAIA exam configuration is 90 questions in 150 minutes across three domains.

What score is required for AAIA?

ISACA uses a scaled score, with 450 on an 800-point scale listed as the passing score in the exam configuration. Practice-test percentages are not equivalent to an official ISACA scaled score.

How should I use AAIA practice questions?

Use mixed questions to find broad knowledge gaps, then use domain practice to review the concepts and explanations behind incorrect answers. FlashGenius readiness thresholds are study guidance, not official ISACA passing scores.

Which AAIA domain has the greatest weight?

AI Operations has the largest listed weight at 46%, followed by AI Governance and Risk at 33% and AI Auditing Tools and Techniques at 21%.

Explore AAIA Tests