CyberCertPrepAI Security FundamentalsAdversarial ML & Model AttacksEASYFree question
A fintech architect is asked to explain the difference between a black-box and a white-box adversarial attack. Which statement is correct?
- A.White-box attacks require physical access to the inference hardware; black-box attacks are remote
- B.In a black-box attack the adversary modifies training data; in a white-box attack the adversary modifies inference inputs
- C.In a white-box attack the adversary has full knowledge of the model (architecture, weights, gradients); in a black-box attack the adversary can only query the model's inputs and outputs
- D.Black-box attacks are always weaker than white-box attacks and pose no real threat to production models
Why C is correct
White-box attacks assume the adversary knows the model architecture, weights, and can compute gradients-enabling gradient-based methods like PGD. Black-box attacks assume only query access (inputs and outputs). Black-box attacks are not always weaker; transfer attacks and decision-boundary attacks can be highly effective. Training vs. inference modification distinguishes poisoning from evasion, not black/white box. Physical access is not required.
Know someone studying for AI Security Fundamentals? Send them this one.