IoT Security in the Age of Large Language Models: Opportunities, Threats, and Challenges Ahead

IoT Security in the Age of Large Language Models: Opportunities, Threats, and Challenges Ahead

azam hokmi1

1) Faculty Member, Department of Computer and Information Technology, Islamic Azad University, Hamedan Branch, Iran Email:

IOI Code : AAHD-ACZBFZ
Publication : National Conference on Information Technology, Nanotechnology, Artificial Intelligence and Technological Futures Studies(itnaf.ir)
Abstract :
    The expansion of the Internet of Things in recent years has presented new challenges to network security. Connected devices are easy targets for attackers due to resource constraints and heterogeneity, and traditional intrusion detection methods, including signature-based approaches and classic machine learning models, have lost their effectiveness against complex and zero-day attacks. In the meantime, large language models with capabilities such as natural language understanding, multi-stage reasoning, and in-context learning have been proposed as a new solution. In this paper, we have reviewed recent research (2024 to 2026) with a review-analytic approach to demonstrate the practical applications of these models in IoT security. The findings suggest that these models have made a difference in four key areas: intrusion detection with 98.2% accuracy (SecurityBERT model), automated analysis of logs and unstructured data, zero-day attack detection with recovery-oriented approaches, and security automation with intelligent agents. However, there are also serious challenges: more than 72% of the research did not investigate the feasibility of running these models on edge devices, 89% of the approaches were not evaluated against competing attacks, and 92% did not provide information on latency and energy consumption. Finally, we propose solutions such as the development of lightweight models, hybrid cloud-edge architectures, federated learning, and standardized evaluation frameworks, emphasizing the need to move from laboratory research to real-world deployment.

Keywords : internet of things large language models cybersecurity intrusion detection in context learning competitive attacks