A pervasive mobile computing environment is typically composed of multiple fixed and mobile entities that interact autonomously with each other with very little central control. Many of these interactions may occur between entities that have not interacted with each other previously. Conventional security models are inadequate for regulating access to data and services, especially when the identities of a dynamic and growing community of entities are not known in advance. In order to cope with this drawback, entities may rely on context data to make security and trust decisions. However, risk is introduced in this process due to the variability and uncertainty of context information. Moreover, by the time the decisions are made, the context data may have already changed and, in which case, the security decisions could become invalid. In this paper, we propose a Monte Carlo based framework to evaluate context based security policies. In this framework, context data is modeled and risk in security decisions is measured.