Adversarial Examples

Adversarial examples are inputs to AI models, especially neural networks, that are subtly altered by an attacker to cause the model to make a confident but incorrect prediction. These small, often imperceptible, perturbations highlight vulnerabilities in a model's robustness.