A Comprehensive Survey of AI Security Threats in IoT and Edge Cloud Systems

Abstract

AI security is a key design criterion for IoT and edge cloud architectures where learning models are embedded near physical processes, receive streaming data from sensors, and are subject to continuous updates. This survey research work proposes a lifecycle-based understanding of AI security threats and proposed a unified taxonomy for the four major categories of threats observed in real-world settings, namely, data poisoning and backdoor attacks on learning model updates, adversarial attacks on model outputs through input manipulation, privacy leakage of confidential information through model-based queries, and model extraction for intellectual property theft and creation of rogue replicas of learning models. The research in surveying these defenses takes account of the limitations posed by the use of edge computing platforms from a computational standpoint, latency tolerance, network connections, and variety of hardware. These defense mechanisms include model provenance and data screening, robust training and backdoor attacks, monitoring and calibration, privacy-preserving learning and access control, and model protection through throttling, fingerprinting, watermarking, and attestation. Unlike prior surveys that treat these threats separately, this survey unifies them under a lifecycle-based taxonomy tailored to IoT and edge cloud deployments and emphasizes deployable defenses under latency, compute, and hardware constraints.