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
Q421 A bank uses a model to forecast loan defaults from historical records. Which model category does this BEST represent?
- A. Predictive model
- B. Reactive machine
- C. Agentic model
- D. Generative model
Answer: A
Predictive models use historical data to forecast outcomes, such as default risk. Generative models create new content, and agentic models take autonomous action toward goals.
Domain 3: AI Technologies and Controls
Q422 A model learns from labeled examples to map inputs to known outputs. Which learning approach does this BEST describe?
- A. Unsupervised learning
- B. Reinforcement learning
- C. Supervised learning
- D. Federated learning
Answer: C
Supervised learning trains on labeled data to predict known outputs. Unsupervised learning finds structure without labels, reinforcement learning uses rewards from interaction, and federated learning trains across decentralized data.
Domain 3: AI Technologies and Controls
Q423 A model groups customers into segments without any predefined labels. Which learning approach does this BEST represent?
- A. Unsupervised learning
- B. Reinforcement learning
- C. Transfer learning
- D. Supervised learning
Answer: A
Unsupervised learning discovers patterns or clusters in unlabeled data. Supervised learning needs labels, reinforcement learning learns from rewards, and transfer learning reuses a pretrained model.
Domain 3: AI Technologies and Controls
Q424 An agent improves its policy by receiving rewards and penalties from the environment. Which approach does this BEST describe?
- A. Supervised learning
- B. Unsupervised learning
- C. Rule-based programming
- D. Reinforcement learning
Answer: D
Reinforcement learning optimizes behavior through reward feedback from interaction. Supervised and unsupervised learning rely on datasets rather than environment feedback, and rule-based systems do not learn.
Domain 3: AI Technologies and Controls
Q425 Which risk is MOST associated with a deep learning model compared with a simple decision tree?
- A. Smaller data requirements
- B. Easier auditing
- C. Reduced transparency and explainability
- D. Lower accuracy on all tasks
Answer: C
Deep learning models are harder to interpret, which makes explainability and audit difficult. They usually need more data and often perform well, so lower accuracy and easier auditing are not the issue.
Domain 3: AI Technologies and Controls
Q426 A team reuses a pretrained model and adapts it to a new task with a small dataset. Which technique does this BEST describe?
- A. Data poisoning
- B. Transfer learning or fine-tuning
- C. Model inversion
- D. Federated averaging
Answer: B
Transfer learning or fine-tuning adapts a pretrained model to a new task. Poisoning and inversion are attacks, and federated averaging combines decentralized training updates.
Domain 3: AI Technologies and Controls
Q427 An LLM answers using documents retrieved from an enterprise knowledge base at query time. Which technique does this BEST describe?
- A. Differential privacy
- B. Retrieval-augmented generation
- C. Adversarial training
- D. Model backdoor
Answer: B
RAG grounds an LLM with retrieved external documents. Backdoors are attacks, adversarial training hardens models, and differential privacy adds statistical noise to protect individuals.
Domain 3: AI Technologies and Controls
Q428 Which control is MOST useful to reduce hallucinated answers from an LLM used for customer support?
- A. Disabling all logging so that prior errors are not reused in new answers
- B. Grounding responses in approved content and requiring human review for high-impact answers
- C. Using older training data so that the model relies on stable facts
- D. Increasing the response length so the model has more room to qualify its claims and caveat uncertain statements
Answer: B
Grounding in approved sources and human review reduces fabricated output. Longer responses and disabling logs do not improve accuracy, and older data may worsen relevance.
Domain 3: AI Technologies and Controls
Q429 The security team wants controls built into an AI product from inception through end of life. Which concept does this BEST represent?
- A. Security by obscurity
- B. Compliance by exception
- C. Reactive patching
- D. Secure by design
Answer: D
Secure by design builds security into development from inception through end of life. Obscurity, exception-based compliance and reactive patching do not embed controls early.
Domain 3: AI Technologies and Controls
Q430 Which is MOST clearly one of the three secure by design principles referenced in the manual?
- A. Take ownership of customer security outcomes
- B. Hide security failures from customers
- C. Delay patches until annual review
- D. Shift all responsibility to users
Answer: A
The principles are to take ownership of customer security outcomes, embrace radical transparency and accountability, and build organizational structures to achieve security goals. The other options contradict them.