Pixels of Deception: How Evolutionary Algorithms Break AI Reliability

The article reviews the evolutionary universal perturbation attack, a technique that uses genetic algorithms to craft tiny, invisible noise patterns that can mislead deep neural networks without needing internal model access.