The objective of the proposed research is to investigate attack and defense measures in deception of learning in control systems. Deception is studied as attacks that manipulate these systems to behave as intended by the adversary. We first study deception from the adversary's standpoint, and compute the optimal attack input to be added to deceive the controller's learning mechanism. This is studied under a non-zero sum game framework.  Secondly, we study deception from the defender's standpoint. In this, we first investigate how, under a Stackelberg game framework, the agent can optimally counteract the adversary's deception. We also develop a novel defense mechanism using `reinforcement unlearning' to remove the deception. We propose to extend our study to other systems such as networked control systems, and other game frameworks.