Scalable and highly available systems often require data stores that offer weaker consistency guarantees than traditional relational databases systems. The correctness of these applications highly depends on the resilience of the application model against data inconsistencies. In particular regarding application security, it is difficult to determine which inconsistencies can be tolerated and which might lead to security breaches. In this paper, we discuss the problem of how to develop an access control layer for applications using weakly consistent data stores without loosing the performance benefits gained by using weaker consistency models. We present ACGreGate, a Java framework for implementing correct access control layers for applications using weakly consistent data stores. Under certain requirements on the data store, ACGreGate ensures that the access control layer operates correctly with respect to dynamically adaptable security policies. We used ACGreGate to implement the access control layer of a student management system. This case study shows that practically useful security policies can be implemented with the framework incurring little overhead. A comparison with a setup using a centralized server shows the benefits of using ACGreGate for scalability of the service to geo-scale.