Large language models (LLMs) that are trained for tasks related to code, such as code generation, do not always perform well at detecting vulnerabilities in code, especially a complex codebase like the Linux kernel, even though they have huge potentials. In this talk, we introduce a framework for finding bugs and analyzing security for the Linux kernel that uses security specific LLMs along with vulnerability data and attack descriptions to improve finding and examining vulnerabilities. We describe how this framework operates on multiple kernel modules and interfaces between kernel and hypervisor like VBS (Virtualization based security). We demonstrate that LLMs need not only reasoning skills, but also deep analyses like data-flow analysis, symbolic execution, and diverse code information to accurately identify different vulnerabilities. We also discuss unresolved challenges towards providing a semi-automatic LLM-assisted Kernel vulnerability detection and patching tools.