Evade AI Model
Evading an AI model is like tricking a security guard by wearing a disguise or showing them a fake ID. Just as a guard might be fooled by a clever costume, an AI can be tricked by subtle changes to data or realistic fakes, causing it to misidentify what it is seeing and let something harmful pass through.
AI model evasion involves manipulating input data to force a model into making an incorrect classification or prediction. This is achieved by crafting adversarial examples—subtle perturbations that cause misclassification—or generating deepfakes that mimic authentic inputs. These techniques are used to bypass security controls like malware scanners or biometric authentication systems, facilitating unauthorized access or malicious activity.
Adversaries can Craft Adversarial Data (AML.T0043) that prevents an AI model from correctly identifying the contents of the data or Generate Deepfakes (AML.T0088) that fools an AI model expecting authentic data. This technique exploits the model's decision boundaries to evade downstream tasks, such as AI-based malware detection or network traffic analysis. Furthermore, deepfake generation serves as a vector for bypassing biometric authentication, enabling initial access by subverting the model's feature extraction and classification logic.