A key feature for most collaborative applications is their ability to distinguish access rights to shared documents in a dynamic group of collaborators. To achieve high availability and avoid single points of failure, systems can replicate this access control policy across sites. Increasingly, Conflict-free Replicated Data Types (CRDTs) are used to this end. They embed conflict resolution strategies that take into account concurrent modifications to the policy but also to the document that it manages. However, few systems in the literature support multiple administrators, and those which do fall short of considering distributed applications where no node or group of nodes is central to its operation. To allow multiple concurrent edits of the policy in massively collaborative settings with no central server, we devise a specialized causal model that minimizes concurrency. Crucially our model allows to correct the document state in case operations have become unauthorized following a policy change. We apply our model to a CRDT replicating an access control policy with read and write rights on a collaborative document and admin rights on the policy. We end by studying flexible conflict resolution strategies, providing our underlying algorithms.