The White-Box Approach to IoT Security: Generating MUD Profiles with RAG

The growing number of Internet of Things (IoT) devices increases the risk of insecure and uncontrolled network communication. The Manufacturer Usage Description (MUD) standard helps reduce this risk by defining network policies for known and legitimate device connections. However, writing MUD profiles manually is time-consuming and often inconsistent, making it difficult for manufacturers to keep up with the scale of IoT development. In this project, we present an automated approach that uses Retrieval-Augmented Generation (RAG) with large language models (LLMs) to generate MUD profiles directly from device documentation and source code. We tested the framework in a simulated network environment that mimics real IoT behavior, extracting relevant communication facts and converting them into a complete and standards-compliant MUD profile. Our results show that this approach can make MUD generation faster, more accurate, and more practical, ultimately improving the security of IoT devices in a simple and efficient way.