Adversarial Attacks

Deliberate, often small, perturbations to input data designed to fool a machine-learning model into making incorrect predictions. They expose vulnerabilities in models and are a key concern for AI robustness and security.