The concept of inference-proofness has been introduced for capturing strong confidentiality requirements—including privacy concerns—of an information owner, communicating with a semi-honest partner by means of their message exchanging computing agents according to some agreed interaction protocols. Such protocols include closed-query evaluation and view generation by the information system agent under the control of the information owner, and the corresponding request preparation by the client agent. The information owner employs a security mechanism for controlled interactions, shielding the epistemic state of the information system agent and suitably altering messages sent to the client agent. The alterings provably guarantee that the partner cannot infer the validity of any piece of information that the information owner has declared as being prohibited. Based on selected previous work, we carefully describe and inspect the underlying function and attack scenario and summarize and analyze basic approaches for controlled interactions within an abstract framework for epistemic states.