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 3: AI Technologies and Controls
Q441 Which activity BEST belongs in the collect and process data phase?
- A. Cleaning duplicates and errors and checking completeness, relevance and timeliness
- B. Retiring the model and archiving its artifacts after final monitoring
- C. Terminating the contract with the model vendor after the pilot ends
- D. Final production monitoring of drift and incident reports for the live model, with results sent to executive management
Answer: A
Collection and processing includes cleaning, completeness and quality checks, and relevance and timeliness evaluation. Retiring and monitoring belong to later phases.
Domain 3: AI Technologies and Controls
Q442 Early risk measurements differ from those measured later in the life cycle. Which explanation fits NIST AI 100-1 BEST?
- A. Risk is always visible at the start, so early measurement is final
- B. Some risk is latent and may increase as AI systems adapt and evolve
- C. Risk only decreases over time as models mature and users learn
- D. Risk never changes after the first assessment, so repeat reviews add cost
Answer: B
NIST notes risk measured early may differ from later measurements because some risk is latent and may grow as systems adapt. Risk is not static, so assessment must continue through each phase.
Domain 3: AI Technologies and Controls
Q443 An AI developer releases a pretrained model and a deployer uses it for a new purpose. Why can their risk views MOST differ?
- A. Risk is identical for both because the model code is the same
- B. Deployers always know more than developers about residual risk
- C. Deployers may not recognize use-specific risk that the developer perceived differently
- D. Developers have no knowledge of the pretrained model they publish, so only the deployer can assess how the model behaves
Answer: C
Different AI actors have different risk perspectives, and deployers may not see use-specific risk that differs from what the developer considered. All actors share responsibility for a trustworthy system.
Domain 3: AI Technologies and Controls
Q444 A model passes accuracy benchmarks but fails when fed unusual edge-case inputs. Which test type would have MOST directly exposed this weakness?
- A. Stress testing against extreme conditions and edge cases
- B. A user satisfaction survey after go-live
- C. Benchmark accuracy testing alone on the validation set
- D. A review of the vendor's marketing claims
Answer: A
Stress testing subjects a model to extreme conditions or edge cases to confirm robustness. Standard benchmarks measure accuracy, precision and recall on expected data, so they can miss unusual inputs. Surveys and marketing reviews do not test robustness.
Domain 3: AI Technologies and Controls
Q445 A team asks how a credit model would behave if unemployment doubled. Which testing approach does this BEST represent?
- A. Data masking
- B. Penetration testing
- C. Code obfuscation
- D. Scenario analysis
Answer: D
Scenario analysis tests the model in hypothetical situations to predict real-world behavior and effectiveness. Penetration testing probes security weaknesses, and obfuscation and masking protect code or data rather than testing outcomes.
Domain 3: AI Technologies and Controls
Q446 Which result of testing MOST appropriately enables a validated AI solution to move into production?
- A. Informal approval from the model developer who built the solution, given by email after a short demonstration
- B. Completion of the marketing launch plan for the new product
- C. Sign-off from the vendor's sales team on the delivered features
- D. Formal authorization under the organization's change management process
Answer: D
Validated solutions should be authorized for production per the change management process, with identified risk placed in a mitigation plan or the risk register. Developer or sales approval does not provide governance. Marketing readiness is not a gate.
Domain 3: AI Technologies and Controls
Q447 Before full roll-out, a company releases an AI tool to a small group in a controlled environment. Which deployment practice does this BEST represent?
- A. Decommissioning
- B. Data masking
- C. Piloting
- D. Model inversion
Answer: C
Piloting tests the solution in a controlled environment before wider deployment. Decommissioning ends a model's life, and inversion and masking are unrelated to deployment.
Domain 3: AI Technologies and Controls
Q448 After deployment, which activity is MOST essential during the operate and monitor phase?
- A. Assessing intended and unintended impacts and tracking performance and drift
- B. Disabling logs to reduce storage costs for the operations team
- C. Ignoring user feedback unless it reaches the executive committee
- D. Rebuilding the model from scratch every month regardless of performance
Answer: A
Operators need mechanisms to assess both intended and unintended impacts, with feedback structures that trigger action by owners. Monthly rebuilds are not required, and disabling logs or ignoring feedback blinds the organization.
Domain 3: AI Technologies and Controls
Q449 What is the MOST important consideration when retiring an AI model?
- A. Leave the model running in production without a named owner
- B. Securely handle the model, training data and logs per retention and destruction requirements
- C. Delete everything immediately without keeping any records of the retirement, to avoid exposing the data at a later date
- D. Publish the model weights publicly so others can maintain them
Answer: B
Decommissioning should address data, models and logs per retention policy, with documentation and communication. Immediate deletion without records or unowned running models creates compliance and security risk.
Domain 3: AI Technologies and Controls
Q450 Before buying a third-party dataset for training, which question is MOST important for the information security team to ask?
- A. How many sales awards the vendor has received in recent years
- B. How colorful and modern the vendor's website and brand design appear
- C. Whether the vendor sponsors the largest industry events each year
- D. Where did the data originate and under what agreement may it be used
Answer: D
Data acquisition should be backed by processing agreements and procedures identifying the origin and permitted use of data. Website design, awards and sponsorship do not address legal or security risk.