Vulnerable AI: A Survey

Abstract

Nowadays, artificial intelligence (AI) is finding relevance in diverse facets of human existence. For example, AI has been applied in areas such as finance, energy, healthcare, transportation, education, agriculture, and entertainment, just to mention a few. However, technologies that employ AI algorithms such as machine learning algorithms and deep learning algorithms are susceptible to a new challenge known as adversarial machine learning (AML) attacks. An AML attack occurs when a malicious actor crafts adversarial examples (AEs) to introduce perturbation into the original input data of an AI-based classification model to cause the model to misclassify inputs. This paper reviewed how computer vision systems, audio systems, industrial systems, transportation systems, blockchain technologies, and cybersecurity systems may be susceptible to AML attacks. Special attention is given to the discussion on AML attacks on network intrusion detection systems (NIDSs) while also highlighting adversarial defense strategies. The paper concludes with a proposal for future research direction regarding AML attack mitigation.