Intent-Driven Access Control List Configuration Based on Large Language Models

The recent advancements in technology, especially in Artificial Intelligence (AI) have tremendously augmented the pace of innovation across various domains. In the past decade or so, AI's incredible powers have simplified several complex tasks, that have greatly reduced time and effort required for manual execution. One such task is the management and configuration of Access Control Lists (ACLs) which can sometimes be time-consuming and prone to human error in large complex network environments. These errors can expose networks to security vulnerabilities. This paper presents a novel approach to automating ACL configuration based on the intent of IT administrators leveraging Large Language Models (LLM) implementation. The proposed system allows administrators to input their desired ACL configuration in plain English, which the LLM then translates into appropriate commands for configuring the ACL. Contrary to generic language models, the authors have created this model from scratch based on GPT architecture specifically by training it on ACL-related data. This tailored approach not only enhances its performance in configuration tasks for ACL but also makes the model less resource-intensive and lighter. Its efficient design allows for easy integration into router firmware ensuring minimal computational overhead. Experimental results demonstrate that the model achieves an accuracy of 90% and its efficiency and practical potential is evident in terms of cost.