What is the term for an attack in which a threat actor deliberately introduces malicious examples into an AI model's training dataset to corrupt its learned behavior?

Question

Accepted Answer

A. Data poisoning. Data poisoning injects malicious training examples to corrupt the model's learned behavior — either degrading overall accuracy or inserting backdoors that trigger specific misclassifications. The SecAI+ objectives include mitigating adversarial risks targeting data pipelines. Model inversion attempts to reconstruct training data from model outputs. Membership inference determines whether a specific record was in the training set. Evasion attacks manipulate inputs at inference time.

What is the term for an attack in which a threat actor deliberately introduces malicious examples into an AI model's training dataset to corrupt its learned behavior?

Related Questions

Discussion