CYBERSECURITY RISKS FROM AI–PHYSICAL CONVERGENCE

Abstract

The primary goal of this research is to identify the extent to which ongoing artificial intelligence integration into physical systems creates new types of multidomain cyber threats. It also aims to illustrate how the combination of AI and physical systems creates unique and unpredictable security risks with the potential to cause physical harm. In order to accomplish these objectives, a conceptual framework based on literature, defense analysis and threat modeling was applied to identify and categorize the major vulnerability areas in AI-physical systems such as systemic vulnerabilities, adverse input vulnerabilities, network intrusion vulnerabilities, and supply chain vulnerabilities. Data collected through case studies and documented events such as Stuxnet and hacking of autonomous vehicle systems, indicated that AI-physical system convergence significantly increases uncertainty and therefore allows for sensor spoofing, model corruption and stealthy adversarial attacks. Additionally, these data indicate that AI-physical system convergence has the potential for causing physical damage and operational failure. Current Information Technology-centric cybersecurity paradigms are inadequate for protecting AI-physical systems, particularly when those systems are used within critical infrastructure or defense applications due to the rapid escalation of attacks and the difficulty of attributing them to a specific party. The convergence of AI and physical systems blurs the distinctions between digital and physical and forces a reevaluation of traditional responses to cyber threats. Therefore, it is recommended that the development of defensive architectures that include multiple layers, real-time anomaly detection, cross-discipline risk assessments, and the design of secure AI models be employed to mitigate the threats associated with AI-physical system convergence. Further, policymakers and industry stakeholders should recognize that cybersecurity cannot be separated from physical safety when dealing with AI-physical systems. Additional recommendations in the study include the development of updated governance models and the inclusion of cyber-physical resilience in military and civilian plans to prevent catastrophic exploitation of AI-based systems.