Anomaly detection and visualization in generative RBAC models

With the wide use of Role-based Access Control (RBAC), the need for monitoring, evaluation, and verification of RBAC implementations (e.g., to evaluate ex post which users acting in which roles were authorized to execute permissions) is evident. In this paper, we aim at detecting and identifying anomalies that originate from insiders such as the infringement of rights or irregular activities. To do that, we compare prescriptive (original) RBAC models (i.e. how the RBAC model is expected to work) with generative (current-state) RBAC models (i.e. the actual accesses represented by an RBAC model obtained with mining techniques). For this we present different similarity measures for RBAC models and their entities. We also provide techniques for visualizing anomalies within RBAC models based on difference graphs. This can be used for the alignment of RBAC models such as for policy updates or reconciliation. The effectiveness of the approach is evaluated based on a prototypical implementation and an experiment.