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
Q691 What is the MAIN reason an organization might need controls for AI beyond existing IT protections such as firewalls and antivirus?
- A. Only vendors need AI controls, not the deploying organization
- B. Existing controls already cover every AI threat completely
- C. AI security is more complex and needs risk-based guardrails
- D. AI systems are never connected to enterprise data
Answer: C
The manual says the complexity of AI controls compared with traditional IT controls requires additional consideration, with guardrails driven by risk assessment or change management outcomes. The organization may augment or implement new controls.
Domain 3: AI Technologies and Controls
Q692 Before choosing controls for an AI initiative, what is the MOST essential factor for applying the correct controls?
- A. Understanding the type of AI being used
- B. Applying the same controls to all AI systems
- C. Choosing the cheapest tool available
- D. Knowing the vendor's marketing strategy
Answer: A
The manual says understanding the type of AI is essential to applying the correct controls, because AI applies to multiple use cases that require different approaches. A single set or cheapest option may leave gaps.
Domain 1: AI Governance and Program Management
Q693 A bank finds its credit model performs worse for one customer group than another. Which action is the BEST first response?
- A. Run a bias audit and examine whether the training data represents that group
- B. Retrain the model on the same data with more epochs
- C. Remove the human appeal process for declined applicants so decisions can be completed without delay
- D. Publish the overall accuracy figure and close the review
Answer: A
A bias audit with a review of dataset representativeness addresses statistical and systemic bias. Reporting average accuracy, removing appeals or retraining on the same data does not fix the disparity.
Domain 2: AI Risk Management
Q694 Which provider activity is MOST closely tied to incident response in the shared responsibility model?
- A. Setting the subscriber's internal access policies and deciding which employees may use the AI system
- B. Choosing which subscribers may view marketing material
- C. Approving the subscriber's data retention schedule
- D. Maintaining an AI incident response plan with staff training and plan testing
Answer: D
Both parties need an AI-centric incident response plan, staff training and testing. Access policies and retention schedules belong to the subscriber, and marketing has no role.
Domain 2: AI Risk Management
Q695 A deployer relies on a vendor's chatbot and wants to limit legal exposure from unauthorized commitments. What is the BEST control?
- A. Remove any disclaimer so customers trust the bot more
- B. Restrict the chatbot to approved policies and add review for commitments it can make
- C. Let the chatbot approve exceptions on its own authority
- D. Assume the vendor will cover every resulting claim because it supplied the model, and skip internal review
Answer: B
The airline example shows a chatbot can bind the deployer as an authorized agent, so the deployer should constrain what the bot can commit to and review. Removing limits or assuming vendor cover leaves liability with the deployer.
Domain 3: AI Technologies and Controls
Q696 Which design choice BEST ensures that a HITL control works as intended in an AI pipeline?
- A. Allow the model to mark its own decisions as reviewed
- B. Require explicit human approval for specified conditions before the workflow continues
- C. Let the pipeline continue automatically if no one responds within the allotted time, to keep throughput high
- D. Display the output to the reviewer after the action is complete
Answer: B
The workflow must be built so that specified conditions require explicit approval before proceeding. Automatic continuation, after-the-fact display or self-approval defeat the control.
Domain 2: AI Risk Management
Q697 Which OWASP LLM risk MOST directly describes vulnerabilities from outdated models, third-party packages or tampered components?
- A. Misinformation
- B. Supply chain
- C. Improper output handling
- D. Excessive agency
Answer: B
The supply chain category covers tampering or poisoning risk through third-party packages, licensing terms, outdated models and similar components. Other categories concern false content, capability and unsafe output use.
Domain 1: AI Governance and Program Management
Q698 A security manager wants AI procedures to reflect enterprise obligations. Which input should shape them MOST?
- A. The default settings of the chosen AI product
- B. The organization's risk appetite and regulatory landscape
- C. The shortest document length that staff will read
- D. The personal preferences of the first team to adopt AI
Answer: B
The manual says enterprise AI policies must reflect risk appetite and the regulatory landscape. Personal preferences, defaults and length do not set obligations.
Domain 2: AI Risk Management
Q699 Which action BEST reduces the risk of unknowingly using pilfered content in an AI-powered product?
- A. Rely on the model's confidence score for originality
- B. Assume public web data is always free to reuse, so no further checks on sourcing or licensing are needed
- C. Remove the contract review to speed up onboarding
- D. Verify the sourcing of the provider's training data and obtain legal review of the contract
Answer: D
Training data from the web may include copyrighted or proprietary content, so sourcing and contract terms need legal review. Confidence scores do not establish originality and web availability does not imply permission.
Domain 3: AI Technologies and Controls
Q700 Which control category in the manual is MOST concerned with evaluating existing access controls against new laws on AI data protection?
- A. Data quality
- B. Access controls
- C. Observability
- D. Configuration management
Answer: B
The manual says existing access controls should be evaluated to meet new regulations and laws protecting data related to AI products and services. The other categories address other aspects.