Security strategies for AI systems in Industry 4.0

Abstract The recent emergence and widespread adoption of Artificial Intelligence (AI) across various industries have not only increased efficiency and innovation but also introduced complex security challenges. Addressing these security challenges is an ever‐evolving area of research that seeks to mitigate issues affecting a wide range of individuals and sectors. Understanding the threats and vulnerabilities posed by these systems, and how to effectively defend against them, has become more crucial than ever. This paper aims to provide an overview of the most common attack vectors and their respective defense strategies, focusing particularly on those relevant to Industry 4.0. A major area of interest is adversarial machine learning, a relatively new field that focuses on corrupting, confusing, and manipulating AI models by intervening in different phases of their life cycle. The key findings indicate that FLAME offers the best protection against data poisoning attacks for neural network image detection models in a federated learning environment. Additionally, due to the specific threat model in Industry 4.0, defensive distillation emerges as the most promising defense strategy against evasion attacks.