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
Q461 Which control BEST protects data integrity in an AI training pipeline?
- A. Colored dashboards that display the status of each job
- B. Longer file names that describe each stage of the pipeline
- C. Hashing, checksums and access logging with change detection
- D. Faster printers for the physical copies of the training reports
Answer: C
Integrity controls such as hashing, checksums, logs and change detection reveal tampering. File names, dashboards and printers do not verify integrity.
Domain 3: AI Technologies and Controls
Q462 Why is backup of AI datasets and models MOST important?
- A. To prevent every incident that could affect the training pipeline
- B. To restore data and models after loss, corruption or attack
- C. To replace access control by keeping a protected copy of every record
- D. To remove the need for encryption of data held in storage
Answer: B
Backups support recovery from corruption, loss or attack, and should be protected like the originals. They do not replace access control or encryption, and cannot prevent incidents.
Domain 3: AI Technologies and Controls
Q463 An organization needs to keep model training data confidential from developers who do not need it. Which control is MOST appropriate?
- A. Storing data on shared drives without controls
- B. Open access for all developers
- C. Role-based access control with need-to-know restrictions
- D. Sharing credentials
Answer: C
Role-based access with need-to-know limits exposure to those who require data. Open access and shared credentials undermine confidentiality.
Domain 3: AI Technologies and Controls
Q464 A customer consented to data use for a service but the company now wants to train an AI model on it. What is the MOST appropriate action?
- A. Delete the data subject's consent record to simplify audits
- B. Use the data anyway because the customer is already a client and the original terms probably imply broad permission
- C. Verify the original consent covers the new purpose or obtain additional consent
- D. Assume the original consent covers every future purpose
Answer: C
Consent is specific to the purpose; if original terms do not cover AI training, additional consent is needed. Using data anyway or assuming broad consent risks violating privacy laws.
Domain 3: AI Technologies and Controls
Q465 A data subject withdraws consent. What is the MOST significant AI-specific challenge?
- A. Removing their influence from a trained model
- B. Updating the logo displayed in the consent management portal
- C. Resetting the password for the data subject's account
- D. Printing a confirmation form for the data subject's records
Answer: A
The right to be forgotten requires deleting data, but removing its influence from an already trained model can be difficult. The other items are trivial.
Domain 3: AI Technologies and Controls
Q466 Which privacy-enhancing technique BEST adds statistical noise so individual records cannot be identified?
- A. Model inversion
- B. Prompt injection
- C. Differential privacy
- D. Data poisoning
Answer: C
Differential privacy adds noise to protect individuals in datasets or outputs. Poisoning, inversion and prompt injection are attacks.
Domain 3: AI Technologies and Controls
Q467 Which technique BEST allows model training across decentralized data without centralizing raw records?
- A. Manual copying of records between regional offices
- B. Data lake aggregation of raw records from all sites
- C. Federated learning
- D. Open sourcing the dataset so others can contribute
Answer: C
Federated learning trains models locally on distributed data and shares updates rather than raw data, reducing privacy exposure. Aggregation in a lake centralizes the data.
Domain 3: AI Technologies and Controls
Q468 An AI model infers sensitive health conditions from innocuous data. Which privacy risk does this BEST represent?
- A. Inference-based privacy risk
- B. Physical access risk
- C. Backup risk
- D. Printer risk
Answer: A
AI can infer sensitive details beyond what is explicitly provided, which is inference-based privacy risk. Physical, printer and backup risks are not inference problems.
Domain 3: AI Technologies and Controls
Q469 Which control is MOST effective to prevent personal data from appearing in a language model's training set?
- A. Adding more GPUs so training completes before leakage occurs
- B. Using larger models with more parameters to improve memory limits
- C. Training for longer so the model generalizes away personal details
- D. Data minimization and scrubbing of personal identifiers before training
Answer: D
Minimizing and scrubbing personal identifiers before training reduces the chance of memorization and disclosure. Bigger models or more compute do not remove personal data.
Domain 3: AI Technologies and Controls
Q470 A privacy officer wants to understand privacy risk before launching a new AI feature. Which control is MOST appropriate?
- A. Rely on vendor assurances only
- B. Wait for complaints
- C. Conduct a privacy impact assessment
- D. Skip documentation
Answer: C
A PIA identifies and mitigates privacy risk before launch. Waiting for complaints or relying solely on vendors leaves risk unexamined.