Leveraging Large Language Models for OT Network Configuration Analysis

Operational Technology (OT) networks face growing cybersecurity risks, yet applying best practice guidelines remains difficult, particularly in settings with limited cybersecurity expertise. This paper proposes a modular framework combining a Large Language Model (Llama3 8B Instruct), semantic search (FAISS), and structured prompting to assist in the analysis of OT configurations. The system extracts best practices from authoritative sources, generates standardized JSON templates for data collection, and leverages a chatbot assistant for compliance validation and mitigation guidance. Experimental results show moderate accuracy (60–66.67%), highlighting both the promise and current limitations of LLM-based security tools. The framework offers a foundation for enhancing automation, interpretability, and resilience in OT environments.