Transforming IoT Security through Large Language Models: A Comprehensive Systematic Review and Future Directions
Mohammed Tawfik, Amr H. Abdelhaliem, Islam S. Fathi · Statistics Optimization & Information Computing · 2025
The rapid integration of Large Language Models (LLMs) in Internet of Things (IoT) security presents both unprecedented opportunities and complex challenges. This systematic literature review examines 34 recent studies (2022-2024) to evaluate the effectiveness, challenges, and architectural innovations of LLM implementations in IoT security environments. Through a rigorous methodology following PRISMA guidelines, we analyze performance metrics, implementation strategies, and resource optimization approaches across diverse security applications. Our findings reveal significant advancements in detection capabilities, with frameworks like SecurityBERT achieving 98.2% accuracy while reducing model size by 89.85%, and privacy-preservation mechanisms demonstrating up to 98.247% protection effectiveness. However, persistent challenges emerge in resource optimization, real-time processing requirements, and cross-platform compatibility. The review identifies critical research gaps in standardization frameworks, ultra-constrained device optimization, and privacy-preserving architectures. Our analysis reveals promising architectural innovations, including hybrid deployment strategies reducing energy consumption by 45% and federated learning approaches achieving 97.12% accuracy while maintaining data privacy. This comprehensive review provides a foundation for future research directions in LLM-based IoT security, emphasizing the need for balanced approaches between security effectiveness and resource constraints. The findings suggest that successful implementation requires careful consideration of computational requirements, privacy preservation, and architectural optimization for resource-constrained environments