A threat actor sends carefully crafted inputs to an AI-based fraud detection model that cause the model to classify fraudulent transactions as legitimate. Which AI-driven attack technique does this describe?

Question

Accepted Answer

A. Adversarial machine learning — adversarial input attack at inference time. Adversarial input attacks craft inputs with subtle perturbations designed to cross the model's decision boundary and produce a misclassification at inference time. The SecAI+ objectives explicitly list adversarial machine learning as an AI-driven threat. Data poisoning occurs during training, not against a deployed model accepting live inputs. Model inversion attempts to reconstruct training data, not manipulate classification outcomes.

A threat actor sends carefully crafted inputs to an AI-based fraud detection model that cause the model to classify fraudulent transactions as legitimate. Which AI-driven attack technique does this describe?

Related Questions

Discussion