Enhancing Smart Home Security with Generative AI : Threat Detection and Countermeasure Design

Smart home environments, while offering convenience through internet-connected devices, are increasingly vulnerable to cyberattacks. Traditional intrusion detection systems are signature-based and often fail to address IoT systems’ dynamic and heterogeneous nature, necessitating advanced security solutions. This thesis proposes a generative AI-based approach to enhance intrusion detection and create adaptive countermeasures in smart home IoT ecosystems. The developed system utilizes a pipeline of AI models: Variational Autoencoders (VAE) for unsupervised anomaly detection, Bidirectional Encoder Representations from Transformers (BERT) for threat classification, and xAI’s Grok3 for generating context-specific countermeasures. Using the AposematIoT-23 dataset, the system is trained to identify anomalous network traffic and device behavior in real time and suggests proactive countermeasures to potential threats. Evaluation results highlight the system’s effectiveness in securing smart home IoT environments. The VAE achieved a recall of 99.2%, ensuring nearly all malicious samples are detected, which is critical for minimizing missed threats in unsupervised anomaly detection. The VAE’s F1-score of 0.95 reflects a balance between high recall and precision (91%), though the subsequent BERT classifier mitigates its false positives. The overall pipeline achieved a 99.88% detection rate for malicious samples with a 0.2%false positive rate in a simulated environment of 10,000 samples. This research underscores the potential of generative AI to significantly enhance IoT security by adapting to novel threats through unsupervised learning for threat detection and providing tailored countermeasures. It represents a step toward more resilient smart home ecosystems, though further optimization is needed to minimize false positives without sacrificing detection capability.