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
Q651 A web application inserts LLM output directly into a database query without validation, and the output contains malicious commands. Which OWASP LLM risk BEST describes this?
- A. Misinformation
- B. Unbounded consumption
- C. Data and model poisoning
- D. Improper output handling
Answer: D
Improper output handling occurs when LLM outputs are not sufficiently validated, sanitized or handled before being passed to other components. Poisoning concerns training data, and the others do not involve passing output downstream.
Domain 2: AI Risk Management
Q652 An attacker alters the data used to train a model so that it contains a hidden backdoor. Which OWASP LLM risk does this MOST directly represent?
- A. Unbounded consumption
- B. Data and model poisoning
- C. Prompt injection
- D. Excessive agency
Answer: B
Data poisoning is manipulation of the data used to train and validate a model to introduce vulnerabilities, backdoors or biases. Prompt injection happens at inference, and the others concern capability and resource use.
Domain 2: AI Risk Management
Q653 Sensitive instructions placed in a system prompt include details that reveal how to bypass guardrails. Which risk has MOST clearly been created?
- A. Supply chain tampering
- B. Unbounded consumption
- C. System prompt leakage
- D. Data and model poisoning
Answer: C
System prompt leakage is the risk that prompts or instructions used to direct the model contain sensitive information such as proprietary details or the means to bypass guardrails. Other risks concern training data, resource use or third-party components.
Domain 2: AI Risk Management
Q654 A retrieval-augmented generation system returns documents a user should not see because the vector store lacks access checks. Which OWASP LLM risk is this MOST closely tied to?
- A. Unbounded consumption
- B. Improper output handling
- C. Vector and embedding weaknesses
- D. Misinformation
Answer: C
Vector and embedding weaknesses in RAG systems can expose unauthorized access, data leaks and manipulated outputs. Misinformation concerns false content and the others involve different failure modes.
Domain 2: AI Risk Management
Q655 After identifying AI-related loss event scenarios, a team wants to complete the identification process. Which activity BEST belongs in that process?
- A. Map the mission and objectives of the AI program and assess the AI impact
- B. Select the controls before any scenarios are written
- C. Document only risks that have already caused incidents elsewhere, since hypothetical scenarios cannot be assessed
- D. Approve all AI projects without further review of the supply chain
Answer: A
The manual lists mapping the mission and objectives, identifying services, products and assets, establishing an AI value chain and assessing AI impact. Approving without review or listing only past incidents is incomplete.
Domain 2: AI Risk Management
Q656 An airline integrates a vendor's chatbot into its customer service, and the bot approves a refund the airline does not allow. Who is PRIMARILY accountable for the consequences under the manual's deployer guidance?
- A. The airline, because the deployer stays responsible in its own environment
- B. The customer, because the bot's reply should have been verified
- C. Neither party, because chatbots cannot enter agreements
- D. The chatbot vendor alone, because it built the underlying language model and controls the training data
Answer: A
The manual says a third-party system does not remove the deployer's responsibility, and the chatbot acted as an authorized agent of the airline. The deployer must manage risk and impact in its deployed environment.
Domain 2: AI Risk Management
Q657 Which statement BEST distinguishes AI provider and AI deployer accountability?
- A. Providers own quality, security and inherent risk; deployers own safe, ethical use
- B. Both roles are identical, so contracts do not need to distinguish them
- C. Deployers carry no responsibility once they buy a licensed product
- D. Providers are responsible for end-user impact, while deployers are responsible for model training
Answer: A
The manual assigns providers the quality, security and inherent risk of their systems and deployers the safe, ethical use and impact on users and stakeholders. The roles differ and an enterprise may be both.
Domain 2: AI Risk Management
Q658 Under the EU AI Act definition used in the manual, which entity is MOST accurately an AI deployer?
- A. An entity that regulates AI use on behalf of a government
- B. An entity that audits vendors but never uses AI itself
- C. An entity that only trains foundation models and sells API access to downstream developers and integrators
- D. An entity that integrates a provider's AI system into its own services
Answer: D
The AI deployer takes an AI system or component from a provider and integrates it into its own processes, products or services. The other roles describe different parties.
Domain 2: AI Risk Management
Q659 A company both builds an AI model and uses it in its own customer service. How should it MOST accurately view its responsibilities?
- A. It acts as both provider and deployer and must meet the responsibilities of each role
- B. It is only a provider because it built the model
- C. It has no external accountability since the model is used internally and no outside party relies on its output
- D. It is only a deployer because it uses the model internally
Answer: A
The manual notes that in some instances an enterprise may be both an AI provider and deployer, so both sets of responsibilities apply. Using the model internally does not remove accountability.
Domain 2: AI Risk Management
Q660 A deployer's AI tool interacts directly with end users and affected communities. Which responsibility is MOST relevant to the deployer?
- A. Managing the impact of the AI system on users and stakeholders
- B. Publishing the vendor's source code to customers
- C. Setting the training schedule for the provider's compute cluster
- D. Defining the inherent architecture of the vendor's foundation model
Answer: A
Deployers often interact with end users and are responsible for safe, ethical use and for managing impact on users and stakeholders. Model architecture and compute scheduling sit with the provider.