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
Q541 Which control is MOST critical in AI change management before a retrained model is promoted to production?
- A. Promotion to production immediately after training has finished
- B. Documented testing, risk review and authorization under the change process
- C. Approval only from the data scientist who retrained the model
- D. Waiting for user complaints to validate that the change is safe
Answer: B
Change management should require testing, risk review and formal authorization, which reduces the chance of unintended behavior. Immediate promotion or self-approval skips independent checks.
Domain 3: AI Technologies and Controls
Q542 A team finishes training a model and wants to move it into production. Which pre-deployment activity MOST reduces the risk of unexpected behavior?
- A. Skipping tests since the training accuracy of the model was high
- B. Allowing users to find defects after release and reporting them
- C. Testing against the original business justification and predefined benchmarks
- D. Testing the model only with the base model before deployment
Answer: C
Validation against the business case and benchmarks confirms the solution meets its purpose. High training accuracy does not prove generalization, and leaving defects to users shifts risk to them.
Domain 3: AI Technologies and Controls
Q543 Which element should MOST importantly be designed into an AI system during model building so incidents can be investigated later?
- A. Storing only final outputs
- B. Automatic recording of key events through logging
- C. Removal of all logging for performance reasons
- D. Manual notes written after incidents
Answer: B
Systems should be designed to record key events automatically so behavior can be reviewed and investigated. Removing logs or relying on later notes loses evidence.
Domain 3: AI Technologies and Controls
Q544 A company retires an AI system. Which activity is MOST important for the training data?
- A. Move the data to personal drives owned by the project team
- B. Keep the data indefinitely in its current location in case it is needed
- C. Publish the data for researchers to use after retirement
- D. Apply retention and secure destruction requirements and document the decision
Answer: D
Decommissioning should handle data per retention and destruction requirements and be documented. Indefinite retention, publication and personal storage increase risk.
Domain 3: AI Technologies and Controls
Q545 A company buys a labeled dataset from a data broker. Which control is MOST important for the data processing agreement?
- A. Omit audit rights so that the broker is not inconvenienced
- B. Allow unrestricted resale of the data to any third party
- C. Define origin, permitted use, quality and breach obligations for the data
- D. Rely on verbal promises from the broker's sales representative
Answer: C
A data processing agreement should state data origin, permitted uses, quality and security obligations. Resale rights, no audit and verbal promises leave gaps.
Domain 3: AI Technologies and Controls
Q546 Which practice BEST ensures that data used for model training stays accurate and relevant over time?
- A. Letting each team maintain its own private copy of the dataset
- B. Deleting the dataset after the first training run is complete
- C. Regular data quality checks and refresh cycles with documented ownership
- D. One-time cleansing before the first model with no later review
Answer: C
Ongoing quality monitoring and refreshing with clear ownership keeps data relevant. One-time cleansing or private copies lead to divergence.
Domain 3: AI Technologies and Controls
Q547 Which technique BEST protects data while it is processed by an AI model in the cloud?
- A. Plain text processing on shared hosts
- B. Disabling encryption in transit
- C. Confidential computing or secure enclaves
- D. Using default passwords
Answer: C
Confidential computing protects data in use through hardware-based trusted environments. Plain text processing, no encryption in transit and default passwords weaken protection.
Domain 3: AI Technologies and Controls
Q548 Data moving between a data lake and a training platform should be protected MOST appropriately by which control?
- A. Using unencrypted file shares that are restricted by IP address
- B. Sending files by email between the lake and the platform
- C. Copying the data through removable media between networks
- D. Encryption in transit such as TLS with authenticated endpoints
Answer: D
TLS with authenticated endpoints protects data in transit. Email, unencrypted shares and removable media risk interception or loss.
Domain 3: AI Technologies and Controls
Q549 A hash of each training file is recorded and verified before each run. Which security property does this MOST directly support?
- A. Non-repudiation of users
- B. Availability
- C. Scalability
- D. Integrity
Answer: D
Verifying hashes detects unauthorized changes, which supports integrity. Availability and scalability are unrelated, and non-repudiation concerns proof of action by a user.
Domain 3: AI Technologies and Controls
Q550 A company wants to use customer data for model training only for the purpose originally disclosed. Which privacy principle does this BEST represent?
- A. Data hoarding
- B. Purpose limitation
- C. Secondary monetization
- D. Open access
Answer: B
Purpose limitation restricts use of personal data to purposes disclosed and consented to. Hoarding, open access and monetization conflict with it.