Why is data integrity important in the context of adversarial risks in AI?

To ensure that training and input data have not been tampered with, which could compromise model behavior

To reduce the size of the dataset for faster training

To ensure the model is deployed on secure hardware

To make the model more interpretable

A financial services company uses a deep learning model to approve credit card transactions. Recently, attackers have been adding small, imperceptible perturbations to transaction data that cause the model to approve fraudulent transactions. The security team wants to reduce the model's susceptibility to such adversarial examples without significantly impacting legitimate transaction approval rates. Which approach should they prioritize?

Use a larger neural network architecture to increase the model's capacity to learn complex patterns.

Implement adversarial training by augmenting the training dataset with adversarial examples generated from the current model.

Apply input sanitization by rounding all transaction amounts to the nearest dollar before feeding them to the model.

Increase the model's input feature space by adding more raw transaction attributes to improve its ability to detect fraud.

Which of the following is a common technique to ensure data integrity in AI systems?

Cryptographic hashing of datasets

A financial institution uses an AI model for credit scoring. The security team must choose a defense against adversarial examples that does not require retraining the model and must not significantly increase inference latency. Which defense mechanism best meets these constraints?

Deploy a separate detector model to identify adversarial inputs.

Use adversarial training to make the model robust to perturbations.

Increase the model's capacity to learn more complex patterns.

Implement input sanitization by normalizing and clipping feature values before feeding them to the model.

Which approach enhances model robustness by reducing the model's sensitivity to small input changes, often by adding a regularization term that penalizes large gradients?

Data augmentation with random crops

CompTIASecAI+

Domain 2Objective 3

Mitigate Adversarial Risks CY0-001 Practice Questions (Page 3)

Part of the Securing AI systems domain, which accounts for 40% of the CY0-001 exam. CompTIA does not publish an official question count, but from its 60-minute exam (~25–40 total, ~10–16 in this domain), expect 3–5 from this objective — we provide 34 practice questions to prepare you well beyond it. (estimate)

34questions here

7free pages

8concepts

40%of the exam

Questions 11–15

11foundation · easy
Why is data integrity important in the context of adversarial risks in AI?
Select an answer first
12application · medium
A financial services company uses a deep learning model to approve credit card transactions. Recently, attackers have been adding small, imperceptible perturbations to transaction data that cause the model to approve fraudulent transactions. The security team wants to reduce the model's susceptibility to such adversarial examples without significantly impacting legitimate transaction approval rates. Which approach should they prioritize?
Select an answer first
13foundation · easy
Which of the following is a common technique to ensure data integrity in AI systems?
Select an answer first
14expert · hard
A financial institution uses an AI model for credit scoring. The security team must choose a defense against adversarial examples that does not require retraining the model and must not significantly increase inference latency. Which defense mechanism best meets these constraints?
Select an answer first
15foundation · easy
Which approach enhances model robustness by reducing the model's sensitivity to small input changes, often by adding a regularization term that penalizes large gradients?
Select an answer first

Finished these 5 questions?

Review the revealed explanations, or continue through the curriculum.

Next 5 questions →Back to CY0-001

Free Basic Practice is a study aid with revealable answers — not a scored exam. Examers.io is independent and not affiliated with or endorsed by CompTIA. “CY0-001” is a trademark of its owner, used for identification only.