Security · Flashcard
What does a training-data poisoning attack do?
- ACorrupts the data a model learns from, so the trained model behaves incorrectly
- BCorrupts the input sent at inference, so a single prediction comes out wrong
- CCorrupts the stored weights after training, so the deployed model is replaced
- DCorrupts the evaluation set, so the model's reported accuracy is overstated
Why this is the answer
Poisoning attacks the learning phase: adversarial examples introduced into the training set become part of what the model believes, and the damage is baked in until retraining. Manipulating a single inference is evasion, swapping stored weights is artifact tampering, and corrupting only the evaluation set misreports quality without changing behaviour.
Official docs