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 AAISM exam practice questions with answers and explanations, organized by domain and part.
Domain 1: AI Governance and Program Management
Q291 Why is data quality MOST important for AI systems?
- A. Data quality matters mainly for traditional systems rather than learning models
- B. Results are only as good as the data the models were trained on
- C. Quality affects only storage cost, not the outputs produced by the model
- D. Models ignore training data and rely only on rules written by developers
Answer: B
AI relies heavily on data, so model performance is often directly proportional to data quality. Models learn from the data, so errors flow into results. Quality affects far more than cost.
Domain 1: AI Governance and Program Management
Q292 A model trained on a dataset with very few minority-class samples produces biased results. Which action is the BEST mitigation?
- A. Train the model for fewer iterations to reduce the effect of the imbalance
- B. Remove all minority-class records so that the remaining distribution is more uniform
- C. Profile the data and apply oversampling, undersampling or cost-sensitive algorithms
- D. Move the dataset to unencrypted storage to speed up preprocessing runs
Answer: C
Data imbalance is mitigated by profiling distributions early and using oversampling, undersampling or cost-sensitive training. Dropping the minority class worsens bias. Fewer iterations or storage changes do not address the imbalance.
Domain 1: AI Governance and Program Management
Q293 An organization has lots of data but little that is consented or licensed for model training. Which problem does this BEST represent?
- A. Data balancing
- B. Data lineage
- C. Data masking
- D. Data scarcity
Answer: D
Scarcity can arise when usable data is limited by consent, licensing, quality, labels or access even where overall data volume is high. Balancing deals with class distribution. Lineage and masking address tracking and protection.
Domain 1: AI Governance and Program Management
Q294 An AI team wants a standard document describing a model's purpose, risk level, training data, bias metrics and contacts. Which artifact BEST fits?
- A. A dataflow diagram
- B. A penetration test report
- C. A model card
- D. A business impact analysis
Answer: C
Model cards document the model's key attributes such as name, purpose, risk level, training data and fairness metrics. Dataflow diagrams show data movement, and penetration reports test security. BIAs evaluate process criticality.
Domain 1: AI Governance and Program Management
Q295 Public LLMs keep evolving and can be jailbroken. Which approach does the manual recommend MOST for AI-generated output?
- A. Trust output once an initial pilot has demonstrated that it performs well
- B. Validate only the outputs that the model itself labels as high risk
- C. Perpetually validate all output through mechanisms to assess and approve it
- D. Rely on vendor assurances about safety testing rather than internal reviews
Answer: C
Because AI is susceptible to manipulation, all output must be perpetually validated, which is the trust-but-verify idea. A successful pilot does not guarantee future behavior. Vendor assurances are insufficient.
Domain 1: AI Governance and Program Management
Q296 An enterprise wants an individual to track AI evolution and document its relationship with AI tools. Which role BEST fits?
- A. An outsourced marketing agency that reports AI trends to the board
- B. An AI lead collaborating with cybersecurity, privacy, legal, procurement, risk and audit
- C. A single developer who documents AI usage independently of other functions
- D. The end user community, which tracks tool changes through informal feedback and reports them to the board
Answer: B
An AI lead should collaborate with a cross-functional group, including cybersecurity, privacy, legal, procurement, risk and audit. Isolation or marketing support would not give the needed perspective.
Domain 1: AI Governance and Program Management
Q297 When calculating the cost-effectiveness of AI, which cost is MOST likely to be overlooked?
- A. The cost of company anniversaries
- B. The cost of printer paper
- C. The cost of office furniture
- D. The cost of security controls
Answer: D
The manual says cost-effectiveness should include the cost of security controls as well as productivity gains and workforce optimization. Office or event costs are marginal. Controls are a real cost of AI.
Domain 1: AI Governance and Program Management
Q298 At what point should an organization FIRST adapt its cybersecurity programs for AI?
- A. Before investing in AI strategies
- B. Only when the vendor insists
- C. Once the AI project is retired
- D. After the first major breach
Answer: A
Adapting or creating AI-centric security practices such as security by design should occur before investing in AI strategies. Waiting for a breach or vendor demand leaves the organization less prepared.
Domain 1: AI Governance and Program Management
Q299 Employees paste proprietary source code into a public chatbot. Which cybersecurity program consideration does this BEST represent?
- A. Threat intelligence
- B. IP leakage
- C. Continuity
- D. Disaster recovery
Answer: B
IP leakage is a key consideration, controlled with permission-based access, visibility tools, and firewall and application controls over where AI tools are used. DR and continuity address resilience. Threat intelligence is about tracking new threats.
Domain 1: AI Governance and Program Management
Q300 A CISO is asked what distinguishes securing AI from AI-enabled security. Which statement is MOST accurate?
- A. Both terms mean the same activity because every security tool contains AI models
- B. Securing AI protects models and data, while AI for security uses AI to improve detection and response
- C. Securing AI refers only to using machine learning to analyze network traffic patterns and to detect anomalies in logs
- D. AI for security means hardening the training data and models against prompt attacks
Answer: B
Securing AI covers protecting the models, data and their intended function, for example against prompt injection, whereas AI-enabled security uses AI to enhance detection, prevention and response. Reversing them is a common error.