We present a game-theoretic model for the spread of deviant behavior in online social networks. We utilize a two-strategy framework wherein each player's behavior is classified as normal or deviant and evolves according to the cooperate-defect payoff scheme of the classic prisoner's dilemma game. We demonstrate convergence of individual behavior over time to a final strategy vector and indicate counterexamples to this convergence outside the context of prisoner's dilemma. Theoretical results are validated on a real-world dataset collected from a popular online forum.