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
Q611 After an AI incident, investigators need evidence of what the model received and returned. Which control would have helped the MOST?
- A. Protected, retained logs of prompts, outputs and model versions
- B. Logging only successful requests, with results reported at the next quarterly meeting
- C. Deleting logs after one day
- D. Allowing developers to edit logs
Answer: A
Protected logs of inputs, outputs and versions provide evidence. Short retention, partial logging and editable logs undermine investigation.
Domain 3: AI Technologies and Controls
Q612 A vendor offers an AI service and refuses to share any security evidence. Which response is the BEST?
- A. Treat it as a risk, require assurance such as audit reports or contract rights, or choose another vendor
- B. Accept it because the vendor is well known
- C. Sign without review to meet the deadline, as most teams already do in practice and in line with how the team normally works
- D. Share more of your own data to build trust
Answer: A
Refusal to provide assurance is a significant risk that should be resolved or avoided. Reputation, deadlines and sharing data do not replace evidence.
Domain 3: AI Technologies and Controls
Q613 A company rolls out a generative AI tool to all staff. Which training content is the MOST important for general users?
- A. How to tune hyperparameters
- B. How to write kernel drivers
- C. What data may and may not be entered, how to verify outputs and how to report problems
- D. How to build a GPU cluster, with results reported at the next quarterly meeting and in line with how the team normally works
Answer: C
General users need acceptable use, verification and reporting guidance. The other topics are for specialists.
Domain 3: AI Technologies and Controls
Q614 Security staff lack skills to test AI systems. Which option is the BEST way to close the gap?
- A. Assume existing penetration testers can cover AI without any learning, so that the team can show early progress to executives
- B. Targeted training and certification, practical exercises and use of qualified partners where needed
- C. Rely on automated scanners only
- D. Delay testing until the skills appear
Answer: B
Targeted training with practice and partners closes capability gaps. Assuming skills exist, delaying or relying on scanners alone leaves AI-specific weaknesses untested.
Domain 3: AI Technologies and Controls
Q615 Management wants to know whether awareness training changed behavior. Which indicator is the MOST useful?
- A. Hours of recorded video uploaded
- B. Number of attendees invited
- C. Fewer policy violations and better results in simulated AI-enabled phishing exercises
- D. Number of emails sent about the training
Answer: C
Behavior-based outcomes show effectiveness. Activity counts show effort, not learning.
Domain 3: AI Technologies and Controls
Q616 A fraud model is stable for months and then declines in performance after a market change. Which control is the MOST effective?
- A. An annual accuracy check
- B. Removing the model's monitoring dashboard
- C. Drift monitoring with thresholds that trigger review, retraining or rollback
- D. Waiting for customers to report errors
Answer: C
Drift monitoring detects changes early and triggers a response. Annual checks and complaints are too slow, and removing dashboards removes visibility.
Domain 3: AI Technologies and Controls
Q617 A team wants to track whether its AI security program is improving. Which set of metrics is the BEST?
- A. Office attendance rates, as most teams already do in practice and in line with how the team normally works
- B. Number of dashboards created
- C. Number of meetings held
- D. Incident counts and severity, time to detect and respond, drift events and control coverage
Answer: D
Outcome metrics show whether risk is being detected and handled. Activity measures do not.
Domain 3: AI Technologies and Controls
Q618 A threat intelligence feed reports a new jailbreak technique against LLMs similar to the one the company uses. What should the team do FIRST?
- A. Ignore it because no incident has occurred yet
- B. Disable the threat feed
- C. Assess exposure, test the application against the technique and update controls if needed
- D. Forward it to marketing
Answer: C
Applying intelligence means assessing exposure and testing, then updating controls. Waiting for an incident or disabling the feed forfeits its value.
Domain 3: AI Technologies and Controls
Q619 An alert indicates that a model's outputs have shifted toward a single outcome class. Which step is the MOST appropriate?
- A. Dismiss the alert because outputs can vary
- B. Increase the alert threshold until it stops
- C. Replace the monitoring tool
- D. Investigate input data and model changes, compare with the baseline and roll back if necessary
Answer: D
A sudden shift needs investigation against the baseline, with rollback if needed. Dismissing or suppressing alerts risks continuing harm.
Domain 2: AI Risk Management
Q620 An attacker slowly feeds crafted examples into an online learning system so that later decisions favor them. Which threat does this BEST describe?
- A. Physical theft of the servers
- B. Denial of service against the hosting network
- C. Data poisoning of the training or feedback data
- D. Phishing of the administrators
Answer: C
Gradual injection of crafted examples to bend future behavior is data poisoning. The other options are different threat types.