KernelGPT leverages large language models to automatically generate high-quality system call specifications for kernel fuzzing, dramatically improving bug detection in operating systems.

• Achieved 8.86x higher code coverage than conventional kernel fuzzing approaches
• Discovered 23 previously unknown bugs in the Linux kernel
• Generated more diverse and semantically valid syscall sequences
• Demonstrated ability to find potentially CVE-worthy vulnerabilities

This research transforms kernel security by automating the traditionally manual process of syscall specification creation, enabling more thorough testing of OS kernels that power billions of devices worldwide.

KernelGPT: Enhanced Kernel Fuzzing via Large Language Models