Improving vulnerability discovery models

Abstract

Security researchers are applying software reliability models to vulnerability data, in an attempt to model the vulnerability discovery process. I show that most current work on these vulnerability discovery models (VDMs) is theoretically unsound. I propose a standard set of definitions relevant to measuring characteristics of vulnerabilities and their discovery process. I then describe the theoretical requirements of VDMs and highlight the shortcomings of existing work, particularly the assumption that vulnerability discovery is an independent process.