Abstract. Inference control can guarantee confidentiality but is costly to implement. Access control can be implemented efficiently but cannot guarantee confidentiality. Hence, it is a natural question to ask when exactly inference control becomes necessary. We characterize the situation in which it becomes possible to infer secrets without any violation of a given access control policy. For this purpose, we establish the Chase as a tool that infers secrets from previous query answers by applying a class of equality- and tuple-generating data dependencies declared over the underlying schema. Our characterization aims to exploit new opportunities for maximizing the availability of data while confidentiality is preserved dynamically and efficiently.