S.T.A.R.K: System for Threat Analysis and Real-Time Kernel-Isolation for IOT Security

In the contemporary landscape of 2026, the exponential proliferation of Internet of Things (IoT) devices within academic and corporate infrastructures has introduced a critical surface for cyber-attacks. Traditional security paradigms, which rely on perimeter-based defenses and reactive patching, are increasingly inadequate against sophisticated firmware-level threats such as hardcoded credentials, hidden backdoors, and insecure binary configurations. This paper presents S.T.A.R.K., an innovative, autonomous zero-trust framework designed to bridge the gap between firmware auditing and real-time network enforcement. The system implements a comprehensive four-stage pipeline: automated binary extraction using Binwalk, deep-layer vulnerability identification, AI-driven risk reasoning via a local Llama 3 Large Language Model (LLM), and autonomous network micro-segmentation. By utilizing a “Never Trust, Always Verify” philosophy, the framework treats every connecting device as a potential threat until its firmware is structurally validated. Experimental results demonstrate that S.T.A.R.K. can successfully identify critical vulnerabilities and trigger an automated “kill switch” using nftables, effectively isolating high-risk devices in a virtualized “digital jail” to prevent lateral movement. This research contributes a scalable, low-latency solution that significantly reduces administrative overhead while fortifying the security posture of modern educational institutions.