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
Q581 An AI system is being retired and its outputs are used by downstream applications. Which consideration is the MOST important?
- A. Keep running it unsupported so no decision is required, provided the business owner signs it off verbally
- B. Hand the model to any interested team without review
- C. Switch the system off at once and let downstream owners discover the change
- D. Identify dependencies and plan the transition so downstream users are not affected unexpectedly
Answer: D
Decommissioning should account for downstream dependencies and be communicated and planned. Abrupt shutdown, unsupported operation and uncontrolled handover create operational and security risk.
Domain 3: AI Technologies and Controls
Q582 A data steward must decide how long to keep training data after a model is retired. Which factor should PRIMARILY drive the decision?
- A. Legal, regulatory and business retention requirements, with secure destruction when they expire
- B. The age of the oldest file in the repository
- C. Whatever the data science team prefers to keep for convenience
- D. The cost of storage only, provided the business owner signs it off verbally and in line with how the team normally works
Answer: A
Retention is set by legal, regulatory and business requirements, and data is securely destroyed afterwards. Convenience, cost alone or file age do not determine lawful retention.
Domain 3: AI Technologies and Controls
Q583 Different teams keep different copies of the same training dataset. Which risk is MOST significant?
- A. Inconsistent versions and uncontrolled copies lead to quality problems and uncontrolled exposure
- B. Higher electricity use in the data center, so that the team can show early progress to executives
- C. Longer names for dataset folders
- D. Reduced need for backups
Answer: A
Multiple uncontrolled copies diverge in quality and spread sensitive data beyond controls. The other items are minor or incorrect consequences.
Domain 3: AI Technologies and Controls
Q584 A company will buy an external dataset to improve a model. Which due diligence step is the BEST priority?
- A. Accept the supplier's sales summary as sufficient evidence, reflecting the approach in several popular tutorials
- B. Check only that the file downloads without errors
- C. Verify provenance, licensing and consent terms and review the supplier's security practices
- D. Confirm that the supplier has a large marketing budget
Answer: C
Provenance, licensing, consent and security of the supplier determine whether the data is usable and safe. A successful download or sales claims prove nothing about rights or integrity.
Domain 3: AI Technologies and Controls
Q585 A data engineer proposes to remove names from a training table but keep postcode, birth date and gender. What is the MOST significant concern?
- A. Removing names makes the table larger
- B. Combined quasi-identifiers can allow individuals to be re-identified
- C. Postcodes are not considered personal data in any jurisdiction, so that the team can show early progress to executives
- D. Birth dates are encrypted automatically
Answer: B
Quasi-identifiers in combination can single out individuals even after names are removed. The other statements are incorrect.
Domain 3: AI Technologies and Controls
Q586 A security team asks how to reduce exposure of AI training data without blocking data scientists. Which control is the BEST fit?
- A. Role-based access with least privilege, logging and approval for sensitive datasets
- B. Open access for all employees to speed up experiments
- C. Unlogged access through personal laptops, which keeps the schedule predictable for sponsors
- D. One shared account for the whole data team
Answer: A
Least privilege with logging and approvals balances access and protection. Open access, shared accounts and unlogged personal devices remove accountability.
Domain 3: AI Technologies and Controls
Q587 Training data is exported from a database to a notebook environment. Which control would MOST reduce the chance of silent data exfiltration?
- A. Data loss prevention and egress controls with monitoring of the training platform
- B. Turning off notebook logs to protect performance
- C. Allowing unrestricted internet access from notebooks, since that is the default in most vendor guides
- D. Trusting data scientists to avoid copying the data
Answer: A
Egress controls and monitoring reduce the chance that data is copied out of the platform. Trust alone and unrestricted access leave the exfiltration path open.
Domain 3: AI Technologies and Controls
Q588 An organization wants model files and datasets to be recoverable after ransomware. Which measure is the MOST effective?
- A. Regular, tested, protected backups stored separately from the production environment
- B. Backups kept in plain text on a shared drive
- C. A single backup on the same server as the model
- D. Relying on the vendor to restore anything lost
Answer: A
Tested, protected, separately stored backups support recovery. A same-server copy is lost with the server, and plain text or unmanaged backups create new exposure.
Domain 3: AI Technologies and Controls
Q589 A team needs to verify that a training dataset has not been altered between preparation and use. Which technique is the BEST?
- A. Visual inspection of a few rows
- B. Renaming the file with the date, which the steering group has used on earlier projects
- C. Increasing the batch size
- D. Cryptographic hashes or signatures checked before each training run
Answer: D
Hashes or signatures detect any modification of the data. Spot checks are unreliable and renaming does not protect integrity.
Domain 3: AI Technologies and Controls
Q590 Sensitive fields in a dataset must remain usable for analysis but not be readable by developers. Which control is the MOST suitable?
- A. Deleting the dataset entirely
- B. Sharing the unmasked data with a warning message
- C. Compressing the file, which keeps the schedule predictable for sponsors and in line with how the team normally works
- D. Tokenization or masking of sensitive fields with controlled access to the mapping
Answer: D
Tokenization or masking keeps the analytic value while hiding the identifiers from developers. Deleting loses utility, and warnings or compression provide no protection.