Adversarial Perturbation Attack

Creates controlled perturbations of known fraudulent transactions to test whether the fraud detection model can be evaded. Operates entirely on tabular transaction features — no documents or media artifacts. The Red Team tests whether plausible feature changes can move transactions across the model's decision boundary.