AI-Driven Secure IoT Onboarding via MUD-based Segmentation in Industrial Networks

The Internet of Things (IoT) introduces significant security risks to local networks. The Manufacturer Usage Description (MUD) standard mitigates these risks by restricting devices to their intended network traffic using Zero Trust whitelists specified by the device manufacturer. However, blindly onboarding MUD files is dangerous. It risks integrating flawed or malicious configurations directly into the local network. This thesis introduces an automated, five-stage pipeline based on related work in the field to analyze MUD files before deployment for their technical readability and file accuracy (correctness and comprehensiveness). The architecture combines deterministic checks, YANG schema validation, and systematic conformance testing, with a vulnerability library ruleset and an AI-driven expert assessment using a Large Language Model (LLM).