A team is testing an image classifier for data poisoning resilience. They have a limited budget and can only run one type of adversarial test. They are most concerned about a backdoor attack that uses a subtle watermark in the corner of images. Which test is most effective?

Inject a set of test images with the suspected watermark pattern and check if the model misclassifies them as the target class

Use a model interpretation tool to visualize which pixels the model focuses on

Run a random label-flipping simulation on the training data and measure the impact on accuracy

Evaluate the model on a large, clean test set to measure overall accuracy

Which of the following is a data sanitization technique to prevent data poisoning?

Removing or correcting suspicious samples before training

Using a larger batch size during training

Increasing the model's learning rate

Deploying the model to a cloud environment

A team is testing a model's resilience to data poisoning. They have a limited budget and can only run one test. They are most concerned about a data injection attack that adds a large number of samples from a new, unseen class. Which test is most effective?

Use a model interpretation tool to see which features are most important

Measure the model's training time to see if it increased due to the extra data

Evaluate the model on a test set that includes samples from the new class to see if it misclassifies them

Run a backdoor trigger test by adding a specific pattern to existing samples

A team has trained a malware detection model. They want to assess its resilience to backdoor attacks before deployment. Which testing approach is most appropriate?

Use cross-validation to ensure the model generalizes well across different data splits

Create a test set that includes samples with known backdoor triggers and verify the model does not misclassify them

Evaluate the model on a standard held-out test set to measure overall accuracy

Monitor the model's inference latency to detect if it is slower on certain inputs

Certified Tester AI Testing

Domain 9Objective 2

Data Poisoning CT-AI Practice Questions (Page 3)

Part of the Domain 9: Methods and Techniques for the Testing of AI-Based Systems domain, which makes up ~13% of our current practice bank. ISTQB does not publish an official question count, but from its 60-minute exam (~25–40 total, ~3–5 in this domain), expect 1–1 from this objective — we provide 27 practice questions to prepare you well beyond it. (estimate)

27questions here

6free pages

7concepts

Questions 11–15

11expert · medium
A team is testing an image classifier for data poisoning resilience. They have a limited budget and can only run one type of adversarial test. They are most concerned about a backdoor attack that uses a subtle watermark in the corner of images. Which test is most effective?
Select an answer first
12foundation · easy
Which of the following is a data sanitization technique to prevent data poisoning?
Select an answer first
13expert · medium
A team is testing a model's resilience to data poisoning. They have a limited budget and can only run one test. They are most concerned about a data injection attack that adds a large number of samples from a new, unseen class. Which test is most effective?
Select an answer first
14foundation · easy
Which type of data poisoning attack involves modifying the correct output label of training samples to a wrong one?
Select an answer first
15application · medium
A team has trained a malware detection model. They want to assess its resilience to backdoor attacks before deployment. Which testing approach is most appropriate?
Select an answer first

Finished these 5 questions?

Review the revealed explanations, or continue through the curriculum.

Next 5 questions →Back to CT-AI

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 ISTQB. “CT-AI” is a trademark of its owner, used for identification only.