Modeling information security policies is an important problem in many domains. Recently, a new access control system, called OrBAC (Organization-Based Access Control) has been proposed. This model brings many solutions to the existing access control systems. However, it does not deal with conflicts due to the joint handling of permission and prohibition policies. This paper deals with the problem of handling conflicts in the OrBAC system, modeled by first order logic knowledge bases. We show that the "blind" application of propositional approaches to inconsistent first order knowledge bases can lead to undesirable conclusions. A solution based on weakening first order formulas responsible of conflicts is proposed.