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
Q411 Which test types should be included FIRST when running adversarial tests against a vendor's LLM?
- A. Token manipulation, gradient descent and jailbreak attacks
- B. Parking audits at the hosting data center campus
- C. Elevator inspections at the vendor's headquarters building
- D. Fire drills and evacuation tests at the vendor's facilities
Answer: A
Adversarial testing includes token manipulation, gradient descent-based attacks and jailbreak attempts. Fire drills and inspections are physical safety exercises.
Domain 2: AI Risk Management
Q412 An organization needs visibility into the components used by a vendor's AI system. Which artifact is MOST useful?
- A. A lunch menu for visiting auditors at the vendor site
- B. A seating chart of the vendor's development team
- C. A software bill of materials
- D. A holiday schedule for the vendor's support staff
Answer: C
SBOMs list the components in a software or AI system, improving supply chain visibility and vulnerability management. The other items provide no component information.
Domain 2: AI Risk Management
Q413 Which party in the AI supply chain is MOST directly responsible for supplying pretrained models and datasets used by others?
- A. Catering vendors that provide meals for development teams
- B. Model and data providers
- C. The deployer's help desk, which handles basic user questions
- D. Janitorial vendors servicing the data center facilities
Answer: B
Supply chain parties include model and data providers, platform and infrastructure providers, integrators and users. Help desk, janitorial and catering vendors do not supply models or datasets.
Domain 3: AI Technologies and Controls
Q414 A system responds to a fixed set of inputs with a predefined output and keeps no memory of past interactions. Which functionality type does this BEST describe?
- A. Reactive machines
- B. Limited memory AI
- C. Self-aware AI
- D. Theory of mind AI
Answer: A
Reactive machines generate a response to a specific, limited set of inputs and have no memory, as in early game-playing systems. Limited memory AI uses recent observations for a task, such as autonomous vehicles. The other two types are theoretical.
Domain 3: AI Technologies and Controls
Q415 An autonomous vehicle uses recent sensor observations to decide its next maneuver but does not retain a long history. Which functionality type does this BEST represent?
- A. Reactive machines
- B. Artificial superintelligence
- C. Limited memory AI
- D. Artificial general intelligence
Answer: C
Limited memory AI uses brief memory of recent observations for specific tasks, which is typical of autonomous vehicles, virtual assistants and recommenders. Reactive machines hold no memory at all, and AGI and superintelligence are not yet realized.
Domain 3: AI Technologies and Controls
Q416 Which category is MOST accurately described as AI that matches human-level ability across many cognitive tasks but does not yet exist in practice?
- A. Limited memory AI
- B. Artificial general intelligence
- C. Reactive machines
- D. Narrow AI
Answer: B
Artificial general intelligence refers to AI with human-level capability across tasks, and remains aspirational. Narrow AI performs specific tasks, which describes today's deployed systems. Reactive machines and limited memory are functional types already in use.
Domain 3: AI Technologies and Controls
Q417 Two neural networks are trained against each other, one creating samples and the other judging whether they are real. Which model type does this BEST describe?
- A. Decision tree
- B. K-means clustering
- C. Linear regression
- D. Generative adversarial network
Answer: D
GANs pair a generator and a discriminator that compete until generated data is hard to distinguish from real data. Decision trees, regression and clustering are not adversarial pairs of neural networks.
Domain 3: AI Technologies and Controls
Q418 What is the PRIMARY role of the discriminator in a GAN?
- A. To store the training data in compressed form for later reuse
- B. To generate new data samples that resemble the training data closely
- C. To distinguish generated samples from original samples
- D. To encrypt model outputs before they are released to end users
Answer: C
The discriminator evaluates whether samples are real or generated, pushing the generator to improve. The generator creates new samples. Storage and encryption are not GAN roles.
Domain 3: AI Technologies and Controls
Q419 A company deploys software agents that make autonomous decisions to maximize supply chain efficiency. Which AI type does this BEST represent?
- A. Agentic AI
- B. Static report generator
- C. Reactive machine
- D. Rule-based macro
Answer: A
Agentic AI can make autonomous decisions and act to achieve specific objectives. Reactive machines only respond to defined inputs, and macros or static reports do not pursue goals autonomously.
Domain 3: AI Technologies and Controls
Q420 Which risk is MOST elevated when an organization deploys generative AI compared with traditional analytics?
- A. Lower regulatory attention because outputs are not personal records
- B. Simpler explainability because outputs are produced in natural language
- C. Reduced data volume requirements compared with traditional analytics
- D. Copyright, intellectual property and harmful content concerns
Answer: D
GenAI raises legal and ethical issues such as copyright ownership, infringement, IP and distribution of harmful content. It generally needs large data volumes and attracts more scrutiny, and explainability is harder, not simpler.