An adversary injects subtle, imperceptible perturbations into medical scan images before submitting them to a diagnostic AI system, causing the system to misclassify cancerous findings as benign. Which attack type does this represent?
Select an answer to reveal the explanation.
Short Explanation and Infographic
Here's the deal — b is correct because adding imperceptible perturbations to inputs at inference time to cause misclassification is the defining characteristic of adversarial example attacks; white-box assumes the attacker has knowledge of the model architecture to craft optimal perturbations. A involves deliberately implanted backdoors in model weights.
Full explanation below image
Full Explanation
B is correct because adding imperceptible perturbations to inputs at inference time to cause misclassification is the defining characteristic of adversarial example attacks; white-box assumes the attacker has knowledge of the model architecture to craft optimal perturbations. A involves deliberately implanted backdoors in model weights. C attempts to reconstruct training data from model outputs. D involves corrupting the training dataset.