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
Q681 In the AI in the loop model, how are humans and AI MOST accurately described?
- A. Humans and AI have identical authority over all decisions
- B. AI offers insights and recommendations while humans keep control of critical decisions
- C. Humans are removed once the AI reaches a target accuracy
- D. AI makes all critical decisions and humans only record the results for audit once the decision has been executed
Answer: B
AI in the loop emphasizes AI assisting and augmenting human decision-making while humans maintain primary control over critical decisions. It is not full automation.
Domain 3: AI Technologies and Controls
Q682 A team deploys a deep learning model that is hard to explain. Which supervision control would BEST help auditors understand how it reached a decision?
- A. Using a separate model to rewrite the outputs without notes
- B. Deleting the logs to limit storage use and reduce the cost of retaining detailed records of each model decision
- C. Detailed logging of decision paths paired with inputs and outputs
- D. Relying on aggregate accuracy reports alone
Answer: C
Complex DL models should be made transparent with detailed logging of decision paths, creating an audit log with the input and inferred output that aids debugging and troubleshooting. Deleting logs or relying on summaries limits visibility.
Domain 3: AI Technologies and Controls
Q683 A team wants to detect early signs that heavy usage may overload an AI system. Which observability category is MOST relevant?
- A. Interpretability techniques such as counterfactual explanations that show which inputs changed the model's output
- B. Adversarial training of the model with crafted examples
- C. Monitoring infrastructure and system health, consumption and processing telemetry
- D. Watermarking the output of the model
Answer: C
Monitoring key system health, consumption and processing telemetry helps identify and anticipate heavy use that could overload resources and affect availability. Interpretability and watermarking address different needs.
Domain 3: AI Technologies and Controls
Q684 A team monitors the pipeline used for training and inference so it can catch unacceptable data quality or prompt injection before they affect decisions. Which observability category is this MOST likely?
- A. Interpretability
- B. Human in the loop
- C. Infrastructure monitoring
- D. Data pipelines
Answer: D
Proactive monitoring of data pipelines for training and inference gives insight into how and why the solution makes decisions and helps prevent issues such as prompt injection or poor data quality. The others cover explainability, resources and oversight.
Domain 3: AI Technologies and Controls
Q685 A complex neural network is difficult to interpret. Which strategy listed in the manual could MOST help improve interpretability?
- A. Removing the audit log to reduce storage
- B. Merging training and validation datasets
- C. Counterfactual explanations
- D. Increasing the number of model layers to improve accuracy
Answer: C
The manual lists counterfactual explanations, feature visualization and influential instance as strategies for improving interpretability of complex models. Adding layers or merging datasets does not help explain decisions.
Domain 3: AI Technologies and Controls
Q686 Which statement about AI observability is MOST accurate?
- A. It is satisfied by publishing a vendor whitepaper, a status page and a summary of the model's accuracy results
- B. It replaces the need for security controls entirely
- C. It supports availability, reliability, performance and trustworthiness
- D. It applies only to the training phase and never to production
Answer: C
The manual says AI observability practices and tools help ensure the availability, reliability, performance and trustworthiness of the system. It complements rather than replaces controls and covers production use.
Domain 3: AI Technologies and Controls
Q687 A security architect reviews how access to AI systems and data is managed. Which control from the manual is MOST appropriate?
- A. Give every user the same access so the model can be tested quickly and permissions are not a bottleneck
- B. Disable recertification because access rarely changes
- C. Adjust access by data and system classification, with MFA and recertification
- D. Allow third parties unrestricted access during development
Answer: C
The manual says access controls should be adjusted by data and AI system classification, use multifactor authentication and include regular monitoring and recertification. Uniform or unrestricted access ignores classification.
Domain 3: AI Technologies and Controls
Q688 A development team wants to reduce the chance that a known vulnerability in its AI environment is exploited. Which control is MOST relevant?
- A. Postponing patches until the next annual review
- B. Disabling automation so changes must be made manually only
- C. Removing the environment from the asset inventory so that scanners and patch tools no longer touch the models
- D. Configuration management that documents, monitors and patches by risk
Answer: D
Configuration management documents, configures by risk, monitors and patches items, and can detect and correct a vulnerability before an attacker exploits it. Automation is encouraged, and delaying patches increases exposure.
Domain 3: AI Technologies and Controls
Q689 A data steward asks what should be documented under the data quality control. Which item BEST belongs there?
- A. The marketing plan for the AI product, including launch dates, messaging and planned customer segments
- B. The cost of storing the data for three years
- C. Only the name of the team that owns the dataset
- D. Quality thresholds and rules, plus how quality is assessed and monitored
Answer: D
The manual says data quality controls should document thresholds and rules, how quality is assessed and issues resolved, and how quality is monitored. Ownership alone or cost items do not meet the control.
Domain 3: AI Technologies and Controls
Q690 The AI change management process is described as critical in the manual. What is its MOST important purpose in security controls?
- A. Remove the need for a risk assessment
- B. Ensure changes and their impacts on the security architecture are considered
- C. Allow developers to modify production without approvals
- D. Speed up releases by skipping testing, since the security architecture rarely changes between versions
Answer: B
The manual says AI change management is critical so that other changes and their impacts on the AI security architecture are considered. It does not replace risk assessment or approvals.