Detection of Adversarial Evasion Attack on AI Model Running on IoT Devices

The attack surface of adversarial is expanding with the wider application of AI-driven IoT devices. It is thus very crucial to research and come up with defense mechanisms. This paper will cover an evasion attack in which the attacker tampers with input and sends the input to bypass detection instead of manipulating the model itself. It provides the unauthorized capacity to carry out data transmission without being detected. We cover the types of adversarial attacks, bring to light the security risks in AI models, and provide a critical examination of existing detection solutions for evasion attacks. Gap analysis is also provided for the revelation of deficiencies in current methods.