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
Q601 Management wants evidence that an AI system can be trusted by regulators. Which approach is the BEST?
- A. Maintain documentation, test results, monitoring records and independent assessments
- B. Provide a single executive statement of confidence
- C. Ask regulators to trust the vendor
- D. Share only the model's accuracy percentage, since that is the default in most vendor guides
Answer: A
Evidence such as documentation, testing, monitoring and independent assessment demonstrates trustworthiness. A statement or one metric does not.
Domain 3: AI Technologies and Controls
Q602 A medical triage assistant occasionally suggests unsafe advice. Which control is the MOST important before wider use?
- A. Clinical validation, safety guardrails and mandatory clinician review for high-risk recommendations
- B. Training the assistant on social media posts
- C. Disabling the review step to improve speed
- D. Hiding the warnings from patients
Answer: A
High-risk health advice needs validation, guardrails and expert review. Removing review, using unvetted data or hiding warnings raises the chance of harm.
Domain 3: AI Technologies and Controls
Q603 A team is designing human-in-the-loop checkpoints. Which criterion is the BEST for deciding where to place them?
- A. Place them wherever reviewers are free
- B. Place them where decisions are high impact, irreversible or affect individual rights
- C. Place them only after harm occurs, which keeps the schedule predictable for sponsors and in line with how the team normally works
- D. Place them on every low-risk formatting step
Answer: B
Human review is most valuable where impact is high or irreversible. Putting it on trivial steps or after harm wastes effort or comes too late.
Domain 3: AI Technologies and Controls
Q604 Reviewers approve 99 percent of AI recommendations within seconds. Which action is the MOST appropriate?
- A. Remove the reviewers to save cost
- B. Congratulate the team for efficiency
- C. Raise the target speed for reviews, because it is the quickest path to the stated goal and in line with how the team normally works
- D. Assess for automation bias by sampling decisions, rotating reviewers and tracking override rates
Answer: D
Rapid near-universal approval may indicate automation bias, so sampling and tracking overrides tests whether oversight is real. Rewarding speed or removing reviewers weakens control.
Domain 3: AI Technologies and Controls
Q605 A company wants a widely used list of attack techniques to guide red team exercises against its AI models. Which source is the MOST suitable?
- A. COBIT governance objectives
- B. MITRE ATLAS
- C. ISO 9001
- D. PCI DSS
Answer: B
MITRE ATLAS catalogs adversary tactics and techniques against AI-enabled systems. PCI DSS, ISO 9001 and COBIT address other topics.
Domain 3: AI Technologies and Controls
Q606 A security architect applies zero trust to an AI platform. Which practice is the MOST consistent with that approach?
- A. Trust any request that comes from the internal network
- B. Grant all developers administrator access
- C. Authenticate users once and never re-verify
- D. Authenticate and authorize every request using identity, device and context regardless of network location
Answer: D
Zero trust verifies each request using identity and context. Trusting the internal network, one-time authentication and broad admin rights contradict it.
Domain 3: AI Technologies and Controls
Q607 An LLM application accepts user text and passes it to internal tools. Which control is the BEST to reduce prompt injection impact?
- A. Remove all logging
- B. Validate inputs and outputs, separate instructions from data and limit tool permissions
- C. Allow the model to run any command it generates
- D. Display system prompts to users for transparency
Answer: B
Input and output validation with permission limits reduce the damage of injected instructions. Running arbitrary commands, exposing prompts and removing logs increase risk.
Domain 3: AI Technologies and Controls
Q608 Developers want to use an open-source model from a public hub. Which check is the MOST important before use?
- A. Rely on the author's profile picture
- B. Use it immediately since it is open source, provided the business owner signs it off verbally
- C. Check how many downloads it has
- D. Verify provenance and integrity, scan the model and dependencies and review the license
Answer: D
Provenance, integrity and scanning reduce supply chain risk, and the license governs use. Popularity and identity signals are not security checks.
Domain 3: AI Technologies and Controls
Q609 An organization discovers staff using an unapproved AI service for client work. What is the MOST appropriate approach?
- A. Ignore it until a breach occurs
- B. Fire everyone involved without review, and later audits can confirm it if needed and in line with how the team normally works
- C. Ban all AI tools permanently
- D. Assess the risk, offer approved alternatives, update the acceptable use policy and monitor
Answer: D
A proportionate response combines risk assessment, approved alternatives, policy and monitoring. Ignoring, punishing without review or blanket bans do not manage shadow AI.
Domain 3: AI Technologies and Controls
Q610 A SOC wants earlier warning of attacks on a model API. Which monitoring signal is the MOST useful?
- A. The size of marketing emails
- B. The color of the user interface, and later audits can confirm it if needed and in line with how the team normally works
- C. Abnormal query volume, repeated near-identical probes and unusual input patterns
- D. The number of chat windows opened
Answer: C
Probing patterns such as heavy or repetitive queries indicate extraction or adversarial attempts. The other items are unrelated to API abuse.