Large Language Model Integration for Intelligent IoT Security and Data Integrity Management

Abstract

Conference Title: 2025 6th International Conference on IoT Based Control Networks and Intelligent Systems (ICICNIS)

Conference Start Date: 2025 Dec. 15

Conference End Date: 2025 Dec. 17

Conference Location: Bengaluru, India

Blockchain networks generate enormous amounts of information of transactions and smart contracts that in most instances are multiplex, multidimensional and difficult to next generation information. Existing analytics methods such as statistics models and machine learning have been left wanting in regards to their semantic dependency capture, situation dependency and novel adversarial behaviour capture in decentralized systems. This is why this work has addressed the use of large language models in blockchain data analysis as one of the ways that can be used to enhance anomaly detection, fraud detection, auditing smart contract, or compliance monitoring. A detailed architecture was introduced that uses a combination of on and off-chain data ingestion, data incorporation through representation learning, and reasoning (by use of LLMs) to learn what can be learned. Experimental findings had suggested that LLMs were proposed to provide better interpretability in transaction monitoring and was better able to provide automated risk assessment than rule-based methods. Scalability, privacy and adversarial vulnerabilities were also of great concern to the paper. The results show that incorporation of blockchain-LLM has the ability to modify features of decentralized governance, regulation adherence, and cross-chain interoperability and that additional research is required to focus on efficiency optimization of models, production of privacy-preserving pipeline, and explainability to trusted implementation of models in real-world on blockchain.