Security and Privacy for Artificial Intelligence: Opportunities and Challenges

The increased adoption of Artificial Intelligence (AI) presents an\nopportunity to solve many socio-economic and environmental challenges; however,\nthis cannot happen without securing AI-enabled technologies. In recent years,\nmost AI models are vulnerable to advanced and sophisticated hacking techniques.\nThis challenge has motivated concerted research efforts into adversarial AI,\nwith the aim of developing robust machine and deep learning models that are\nresilient to different types of adversarial scenarios. In this paper, we\npresent a holistic cyber security review that demonstrates adversarial attacks\nagainst AI applications, including aspects such as adversarial knowledge and\ncapabilities, as well as existing methods for generating adversarial examples\nand existing cyber defence models. We explain mathematical AI models,\nespecially new variants of reinforcement and federated learning, to demonstrate\nhow attack vectors would exploit vulnerabilities of AI models. We also propose\na systematic framework for demonstrating attack techniques against AI\napplications and reviewed several cyber defences that would protect AI\napplications against those attacks. We also highlight the importance of\nunderstanding the adversarial goals and their capabilities, especially the\nrecent attacks against industry applications, to develop adaptive defences that\nassess to secure AI applications. Finally, we describe the main challenges and\nfuture research directions in the domain of security and privacy of AI\ntechnologies.\n