John Homer Promotion Portfolio

Mitigation of security risk is an important task in enterprise network security management. However it is presently a skill acquired by individual experience, more an art than a science. The biggest challenge in the problem is a quantitative model that objectively measures the likelihood a breach can be accomplished. This paper presents a sound and practical approach to such a quantitative model. We utilize existing work in attack graphs and individual vulnerability metrics, such as CVSS, and apply probabilistic reasoning to produce a sound risk measurement. The problem requires a careful coordination of attack graph data to account for cyclic and shared dependencies. We recognize that networks commonly have many host interconnections and network privileges could be gained in many ways. This factor leads to cycles in an attack graph, which must be identified and properly treated when measuring risk to prevent distortion of the results. We also recognize that multiple attack paths leading to the same network privilege will often share some dependencies and so a valid assessment cannot simply treat these paths as independent. Our approach is provably sound and ensures that shared dependencies have a proportional effect on the final calculation, and that cycles are handled correctly so that privileges are evaluated without any self-referencing effect. We also present preliminary experimental results on our algorithm and identify directions for future improvement.