DEFENSE-IN-DEPTH MODEL OF COUNTERMEASURES AGAINST ADVERSARIAL AI ATTACKS: LITERATURE REVIEW AND CLASSIFICATION

The proliferation of artificial intelligence (AI) applications in mainstream businesses has led to a substantial rise in the threat of adversarial artificial intelligence (AAI) attacks. Consequently, it becomes imperative to devise effective countermeasures to mitigate such risks. While the research community has made progress in developing specific countermeasures and controls, a comprehensive synthesis of existing literature, providing an overarching perspective on safeguards against AAI attacks, has been lacking. This paper aims to bridge that gap in the scholarly discourse by presenting a holistic view of countermeasures against AAI attacks. Further, the paper employs a systematic classification of identified countermeasures into three categories: preventive, detective, and corrective controls, based on the defense in depth (D-i-D) model. This framework offers valuable insights for cybersecurity managers, auditors, leaders overseeing AI technologies, and researchers. Our findings reveal a significant emphasis on the development of automated preventive and detective controls to counter AAI attacks. However, there remains a need for further research on procedural or process-based controls and regulatory compliance to enhance the resilience of AI systems.