Evasion attacks occur when an attacker modifies input data (such as adding subtle noise to an image) to trick a model into misclassification. The AAIA™ manual identifies " Adversarial Training " as a primary defense, where the model is intentionally exposed to adversarial examples during the training phase to improve its robustness and resilience. This allows the model to learn the patterns associated with malicious inputs. While static filtering (Option A) and ensembles (Option C) can provide layers of defense, they are often bypassed by sophisticated attacks. Regular bias reviews further ensure that the model's decision-making remains fair and consistent across all inputs, including those designed to exploit algorithmic weaknesses.