The need to develop user adaptive systems have been made more evident by emerging work that applies user modeling to technologies such as automatic interface/software customization and health care support systems. Since different people prefer different styles of interaction, intelligent systems can be designed to adapt to the user's changing needs and preferences. The sequential nature of human-computer interaction makes it possible to build user adaptive systems that learn the user's preferences overtime. However, the main bottleneck in building effective adaptive systems is that adaptive behaviour may cause disruption to the user's mental model of the application. In this work, we propose to model disruption using a dynamic Bayesian network in order for the system to capture the uncertainties and interaction dynamics in a principled manner. Our goal is to build an assistive agent that tailors to individual needs and preferences. We focus on the development of the disruption model for menu adaptations here, but the general principles apply more broadly.