ISACA · Free Practice Questions · Last reviewed May 2026
18 real exam-style questions organised by domain, each with the correct answer highlighted and a plain-English explanation of why it's right — and why the others are wrong.
When setting KPIs for an AI risk program, which metric is a leading indicator of potential model bias?
Average model inference latency
Percentage of customer complaints regarding AI decisions
Number of model updates per quarter
Demographic parity ratio in training data samples
Demographic parity ratio measures outcome rates across protected groups in sampled training data, so an imbalanced ratio precedes biased predictions. It is a leading indicator because it flags imbalance before training, unlike lagging production fairness metrics.
An organization is establishing an AI governance framework. Which approach is most effective for aligning AI risk appetite with enterprise-wide risk management (ERM)?
Map AI-specific risk metrics to existing enterprise risk categories like operational, financial, and reputational risk.
Mapping AI risk metrics onto existing enterprise risk categories lets AI appetite be expressed in the same taxonomy ERM already aggregates and reports. This satisfies the alignment requirement by integrating AI exposure into established operational, financial, and reputational reporting rather than running a parallel framework.
Establish a top-down mandate that ignores business unit input to ensure uniformity.
Focus solely on technical model performance metrics to define the organizational risk appetite.
Create a separate AI-specific risk register that operates independently of the corporate risk register.
The AI Risk Governance Committee is defining the risk appetite for a new generative AI chatbot. Which THREE factors must be considered to align with organizational risk tolerance?
The speed of the development team's sprint cycles
Data privacy and residency requirements for training sets
Data privacy and residency requirements constrain which training data may lawfully be used and where it resides. For a generative chatbot, these legal and contractual limits directly shape acceptable risk exposure, so the committee must weigh them against organisational risk tolerance.
The likelihood of model drift impacting output accuracy
Model drift erodes accuracy as real-world data diverges from training distributions, so its likelihood determines how much output degradation the organisation will tolerate. This directly informs the chatbot's risk appetite, monitoring frequency and retraining triggers.
The impact of potential hallucination on brand reputation
Hallucination risk directly shapes brand exposure, so its potential reputational impact must be weighed against the committee's stated risk tolerance. This factor translates an abstract appetite into a concrete harm threshold, ensuring the chatbot's deployment stays within acceptable organisational limits.
The total cost of GPU infrastructure
A firm is using the NIST AI Risk Management Framework to document its AI risk program. Which approach best ensures that AI risk metrics are dynamic?
Conducting annual risk assessments
Setting static thresholds for model accuracy
Limiting AI deployment to offline systems
Integrating model performance metrics into operational dashboards
Feeding model performance metrics into operational dashboards keeps AI risk measurement continuous rather than point-in-time, satisfying the stem's demand for dynamic metrics. Live telemetry surfaces drift, degradation and emerging harms as they occur, enabling the Govern and Measure functions of the NIST AI Risk Management Framework to trigger timely response.
An enterprise is establishing its AI Governance Committee. Which stakeholder is most critical to include to ensure alignment between AI technical capabilities and the organization's enterprise risk appetite?
Head of Legal
Chief Information Security Officer
Chief Risk Officer
The Chief Risk Officer owns the enterprise risk appetite and translates it into tolerances. Including them ensures AI initiatives are governed against that appetite, aligning technical capability with organisational risk limits rather than leaving risk oversight to engineering alone.
Lead Data Scientist
A cross-functional committee is evaluating the deployment of a high-impact AI model. Which TWO factors are critical to include in the initial AI risk assessment to ensure comprehensive coverage?
The frequency of software patching for the underlying cloud infrastructure.
The socio-technical implications and potential impact on protected groups.
High-impact AI can entrench discrimination, so the assessment must examine socio-technical implications and effects on protected groups. This satisfies the stem's demand for comprehensive coverage by addressing societal harm and fairness, not merely technical performance or narrow accuracy metrics.
The underlying training data lineage, including provenance and potential bias sources.
Bias originates upstream, so tracing training data lineage — provenance, collection methods and transformation history — exposes embedded bias sources. This satisfies the comprehensive-coverage requirement by grounding the risk assessment in the model's actual data foundations rather than only post-deployment symptoms.
The name of the vendor providing the cloud compute resources.
The specific programming language used for the model development.
Want more AI Risk Program Management practice?
Practice this domainYou are mapping AI risks to the NIST AI RMF. Which step is essential when documenting the 'Map' function for a high-risk autonomous system?
Identifying and documenting the context and intended use
The Map function establishes context: intended purpose, users, affected populations, operating environment and constraints. Documenting these for a high-risk autonomous system frames all subsequent risk identification and measurement, ensuring later Assess and Manage activities target the actual deployment scenario.
Updating the SOC2 Type II report
Automating the model deployment pipeline
Conducting a quarterly penetration test
A firm is establishing an AI Governance Committee. Which TWO groups should be represented to ensure comprehensive oversight?
Data Privacy Office
The Data Privacy Office contributes expertise on personal data handling, consent, retention and cross-border transfer, ensuring AI initiatives satisfy privacy regulation. Their inclusion satisfies comprehensive oversight by covering data-subject impact and privacy risk, which technical and business representatives alone would not adequately assess.
Public relations
Legal and Compliance
Legal and Compliance supplies regulatory interpretation across jurisdictions, contractual obligations and liability exposure, ensuring the committee's oversight addresses enforceable requirements. Their representation satisfies the stem's comprehensive oversight constraint by embedding statutory and ethical risk assessment into AI decisions before deployment.
Hardware vendors
External marketing agencies
A financial institution is integrating AI risk into its ERM framework. Which action most effectively aligns AI risk appetite with enterprise risk appetite?
Defining quantitative risk thresholds in the AI governance policy
Defining quantitative thresholds in the AI governance policy directly satisfies the stem's alignment constraint: it translates the enterprise risk appetite statement into measurable AI tolerances, so AI initiatives are approved or escalated against the same monetary and likelihood limits used across other risk domains.
Purchasing cybersecurity insurance for AI systems
Assigning AI risk ownership to the IT department
Implementing a standalone AI audit program
During a board meeting, the Chief Risk Officer needs to present AI model performance trends. Which metric is most critical for board-level reporting?
Data scientists' training hours per quarter
Computational resource utilization percentage
Model latency in milliseconds
Frequency and severity of model drift incidents
Drift incident frequency and severity translate technical model degradation into business risk terms boards govern: how often performance decays and how much harm results. This quantifies operational stability and oversight effectiveness better than granular statistical metrics.
When integrating AI risk into existing ERM, which THREE aspects of an AI model lifecycle must be audited to ensure compliance?
Personnel physical security badges
Office climate control settings
Model validation results
Model validation results evidence that the model meets defined performance, fairness and robustness thresholds before deployment. Auditing these records confirms independent testing occurred, satisfying the lifecycle compliance requirement for AI risk integration within enterprise risk management.
Incident response logs
Incident response logs must be audited because they evidence how AI-related failures, drift or misuse were detected, triaged and remediated. Within ERM, these records demonstrate that model incidents feed back into risk registers and control testing, satisfying compliance and continuous-improvement obligations.
Data lineage and provenance
Data lineage and provenance must be audited because AI compliance depends on tracing training, validation and inference data back to its sources, confirming lawful collection and detecting drift or contamination. Without documented lineage, model outputs and regulatory attestations cannot be substantiated during ERM review.
Which TWO of the following governance actions should a Chief Risk Officer (CRO) implement to align AI risk with the COSO Enterprise Risk Management (ERM) framework?
Establish a cross-functional AI Risk Committee
COSO ERM requires governance and oversight structures spanning the entity. A cross-functional AI Risk Committee embeds AI risk into that oversight layer, giving the CRO a formal body to assign accountability, escalate issues and integrate AI into existing ERM reporting lines.
Assign AI risk ownership to individual data scientists
Disable all open-source AI libraries
Mandate daily model training cycles
Define AI-specific risk appetite statements within the ERM policy
COSO ERM demands risk appetite be defined and communicated. AI-specific risk appetite statements translate the enterprise appetite into measurable AI tolerances, letting the CRO align AI risk acceptance with the ERM policy's objectives rather than treating AI as an exception.
Want more AI Risk Governance And Framework Integration practice?
Practice this domainDuring the AI lifecycle, when should a 'Data Quality' assessment be performed to minimize long-term risk?
After model deployment in production.
During the final model sign-off.
During the decommission phase.
During the data ingestion and preprocessing stage.
Data quality assessment belongs at ingestion and preprocessing, where source data is profiled, cleaned and validated before training. Catching missing values, bias or schema drift here prevents contaminated datasets propagating through model development, satisfying the minimise-long-term-risk constraint.
Which AI lifecycle stage is most susceptible to the risk of 'data leakage' where training data inadvertently contains information from the future/target?
Model monitoring phase.
Model decommission phase.
Deployment phase.
Data preparation and training phase.
Data preparation and training is where datasets are assembled, split and fed to models. If temporal ordering or target variables leak across train and test boundaries here, future or label information contaminates training, making this stage the primary leakage point.
A practitioner is setting up a model monitoring service in AWS SageMaker Model Monitor. They observe that the ground truth data is significantly delayed. What action ensures risk identification remains effective?
Configure Data Quality Monitoring on input features only.
Delayed ground truth defeats real-time label comparison, so Data Quality Monitoring on input features detects distribution drift and anomalies in the incoming data itself, satisfying the need for effective risk identification without waiting for labels.
Disable monitoring until ground truth is available.
Manually force a model redeployment.
Reduce the sampling rate of the inference logs.
You are managing a model that utilizes 'Online Learning'. What is the most critical risk requiring constant lifecycle vigilance?
Hardware failure in the training cluster.
Adversarial data poisoning and instability.
Online learning updates weights continuously from live input, so attackers can inject poisoned samples that gradually corrupt the model, and rapid feedback loops can trigger runaway instability. Constant vigilance is required because no static validation checkpoint exists to catch either failure.
Data drift from batch updates.
Memory exhaustion due to high throughput.
A financial institution uses an AI model to approve loans. After deployment, they notice the model is rejecting loan applications from a specific region at a rate 30% higher than historical human benchmarks. What is the most appropriate next step in the risk lifecycle?
Ignore the shift as it may be a temporary market trend.
Conduct a bias investigation by comparing input distribution shifts between training data and current inference requests.
Comparing training versus inference input distributions isolates whether the elevated regional rejection rate stems from data drift rather than genuine risk factors, directly addressing the 30% disparity against human benchmarks before any retraining or deployment change.
Immediately shut down the system.
Train a secondary model to override the primary one.
In the context of the Google Cloud Vertex AI Model Registry, what is the best approach to mitigate the risk associated with a model update that performs poorly on edge cases?
Implement traffic splitting with a shadow deployment.
Shadow deployment mirrors live traffic to the updated model without serving its predictions, so poor edge-case behaviour is observed in production conditions while users still receive the stable model's responses. This satisfies the constraint of mitigating risk before any traffic shift.
Increase the training epoch count for the new version.
Directly replace the active production model version.
Update the training data to remove edge case records.
Want more AI Lifecycle Risk Management practice?
Practice this domainThe vendor does not publish an exact question count for the AAIR. Always confirm current details on the vendor's official exam page.
Scenario-based questions covering exam objectives with detailed answer explanations.
The exam covers 3 domains: AI Risk Program Management, AI Risk Governance And Framework Integration, AI Lifecycle Risk Management. Questions are weighted by domain — higher-weight domains appear more on your actual exam.
No. These are original exam-style practice questions written against the official ISACA AAIR exam objectives. They are not copied from the real exam. Courseiva focuses on genuine understanding, not memorisation of braindumps.
Courseiva tracks your accuracy per domain and routes you toward weak areas automatically. Free, no account required.