Opacity, a confidentiality property, is an increasing concern in cyber-physical systems (CPSs) that are vulnerable to an intruder which intends to reveal a & x201C;secret & x201D; of a system. This note presents a new framework for opacity and considers the problem of enforcing opacity in CPSs modeled as linear time-invariant systems. The confidential information involves the CPS & x2019; interference attenuation capacity (IAC), called the secret. A system is opaque if the intruder never infers the secret IAC from an observation mapping of system output. To guarantee the confidentiality requirement, an effective algorithm is proposed for synthesizing opacity-enforcing controllers by using a new approximation-based Q-learning. A main advantage of this method is that it does not require any knowledge of the system dynamics matrices. A simulation example is presented to sustain the theoretical results.