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 2: AI Risk Management
Q621 A competitor sends thousands of API queries to approximate a proprietary model's behavior. Which control is the MOST effective?
- A. Publishing the model weights
- B. Rate limiting, query monitoring and access controls on the model API
- C. Increasing the number of free queries
- D. Removing authentication
Answer: B
Limiting and monitoring queries makes extraction harder. Publishing, removing authentication or granting more queries helps the attacker.
Domain 2: AI Risk Management
Q622 Attackers generate a realistic video of the CFO instructing a transfer. Which control is the MOST appropriate?
- A. Disabling transfer logs
- B. Trusting video calls as proof of identity
- C. Allowing any employee to approve transfers, which the steering group has used on earlier projects
- D. Out-of-band verification and approval procedures for high-value requests
Answer: D
Independent verification of high-value instructions defends against deepfake impersonation. Trusting video or loosening approvals invites fraud.
Domain 2: AI Risk Management
Q623 A company wants to reduce hallucination harm in a customer-facing LLM. Which mitigation is the BEST combination?
- A. Grounding in approved sources, output validation and human escalation for sensitive answers
- B. Increasing the response length
- C. Using only older training data, which the steering group has used on earlier projects and in line with how the team normally works
- D. Disabling feedback buttons
Answer: A
Grounding, validation and escalation reduce the impact of fabricated answers. Longer output, no feedback and older data do not.
Domain 2: AI Risk Management
Q624 Leadership accepts a residual AI risk after mitigation. What is the MOST important governance step?
- A. Let the vendor accept it on the organization's behalf
- B. Document the acceptance with an accountable owner and review it periodically
- C. Hide it from the risk register
- D. Leave the decision undocumented, as long as the budget for the work stays unchanged
Answer: B
Risk acceptance should be recorded with an accountable owner and reviewed. Undocumented or hidden acceptance weakens governance, and vendors cannot accept the organization's risk.
Domain 2: AI Risk Management
Q625 A provider updates its model without notice and the deployer's results change. Which control would have MOST helped?
- A. Assuming updates never affect behavior
- B. Contractual notification and change-management terms with testing before adopting updates
- C. Relying on verbal updates
- D. Disabling all updates permanently, with results reported at the next quarterly meeting and in line with how the team normally works
Answer: B
Contract terms requiring notice, with testing before adoption, manage provider changes. Assuming no effect or freezing updates forever are impractical.
Domain 2: AI Risk Management
Q626 A company wants better visibility of components in a third-party AI system. Which artifact is the MOST helpful?
- A. A brochure of customer logos
- B. A map of the vendor's offices, since that is the default in most vendor guides
- C. A list of the vendor's employees
- D. A bill of materials listing models, data sources and software dependencies
Answer: D
A bill of materials identifies components and dependencies for risk assessment. The other items do not.
Domain 1: AI Governance and Program Management
Q627 A sponsor asks why the AI security program needs stakeholder support. Which reason is the MOST accurate?
- A. Support is needed only from IT staff, with results reported at the next quarterly meeting
- B. Stakeholders only approve the budget and do nothing else
- C. Stakeholder support removes the need for controls
- D. Funding, adoption and cooperation depend on leaders understanding the risk and value
Answer: D
Leadership and stakeholder support secures resources, adoption and cooperation. It does not replace controls and extends beyond IT.
Domain 1: AI Governance and Program Management
Q628 An AI model is suspected of leaking personal data in responses. What is the BEST immediate action under the incident plan?
- A. Keep the feature running to avoid disruption, with results reported at the next quarterly meeting
- B. Publicly announce the cause before investigating
- C. Contain the issue by limiting access or disabling the feature while preserving evidence
- D. Delete the logs first
Answer: C
Containment with evidence preservation limits harm and supports investigation. Continuing, deleting logs or premature announcements worsen outcomes.
Domain 3: AI Technologies and Controls
Q629 A team wants to compare two candidate models for a regulated decision. Which criterion should carry the MOST weight?
- A. Ability to meet explainability, fairness and audit requirements with adequate performance
- B. Which model has the best marketing, because it is the quickest path to the stated goal and in line with how the team normally works
- C. Which model has the most parameters
- D. Which model was released most recently
Answer: A
For regulated decisions, compliance with explainability, fairness and audit requirements outweigh size, novelty or marketing.
Domain 3: AI Technologies and Controls
Q630 A company must decide who is accountable for the quality of a training dataset. Which assignment is the BEST?
- A. No one, since data belongs to everyone
- B. The vendor of the model, which keeps the schedule predictable for sponsors and in line with how the team normally works
- C. The newest team member
- D. A named data owner or steward accountable for quality, access and lifecycle decisions
Answer: D
A named owner or steward gives accountability for quality, access and lifecycle. Diffuse or misplaced ownership leads to unmanaged data.