Many languages provide mechanisms for programmers to declare abstract data types (ADTs), hide the details of their implementation, and allow manipulation of these ADTs only by controlled interfaces. This information hiding strategy allows the implementation of the ADT module to be changed without disturbing the client programs.  In Erlang programs structural information about ADTs is exposed by pattern matching and type inspecting built-ins, making it very hard to guarantee that changes in the ADT's implementation will not have devastating effects on client's code.  We have recently extended Erlang with the ability to declare opaque terms (i.e., terms whose structure should not be inspected outside their defining module) and detect violations of their opaqueness using Dialyzer. In this talk we will present this addition to the language and its capabilities, and will show interesting examples of code that (erroneously) depended on implementation details of commonly used library modules (ETS tables, gb_sets, gb_trees, etc.).