thinQmaster Virtual testing partner Adaptive practice exams that find your weak spots and drill them. Every answer explained. From $69
Enroll

thinQmaster

Find your exam

Five questions free, before you decide.

Step 1 · choose a vendor

In one of our camps? Your practice exams come with the tuition — they are on your study page already.

Commonly confused · AI & Machine Learning

Data Poisoning vs Evasion Attack: What's the Difference?

Data poisoning corrupts the training data so the model learns wrong behavior or a hidden backdoor. An evasion attack leaves the model alone and instead crafts malicious inputs at inference time, such as adversarial examples, that cause misclassification.

Data Poisoning

Full definition
vs

Evasion Attack

Data PoisoningEvasion Attack
Lifecycle stageTrainingInference/deployment
What is manipulatedTraining datasetInput sent to the deployed model
ExampleInjecting mislabeled samples so malware is learned as benignAdding imperceptible noise so a stop sign is read as a speed limit sign
PersistencePersists in the model until retrainedPer input
DefensesData provenance, validation, anomaly detection on training dataAdversarial training, input preprocessing, ensemble models
MITRE ATLASPoison Training DataEvade ML Model
Remember it

Poisoning taints the recipe; evasion tricks the taster.

Exam tip

If the attack happens before or during training, it is poisoning. If a deployed model is fooled by specially crafted input, it is evasion. Model inversion and extraction are separate privacy and theft attacks.

Tested on

More AI & Machine Learning look-alikes