This article addresses the problem of escalated violence against minorities induced by interventions. The conventional interventionist wisdom rejects the very possibility of such escalation, claiming that violence against minorities is a monotonically decreasing function of interventions and insisting that it is third-party inaction that is the problem. In recent years, this position has been challenged by scholars, who have argued that the threat of intervention generates a moral hazard, providing minorities with perverse incentives to provoke the very violence the threat of intervention is supposed to deter. I argue that while many of the criticisms of the interventionist position contained in this argument are correct, it is nonetheless not fully convincing, because it fails to explain why the same threat of intervention that radicalizes minorities does not make the states that are the potential targets that much more moderate. Nor does it explain adequately why interveners are unable to make the threat of intervention conditional on the minorities’“good” behavior. I propose an alternative theory of escalation, which depicts third-party interventions as instances of incomplete information bargaining, demonstrating that escalation is a function of certain distributions of private information about the target-state’s level of brutality and the third party’s resolve, as well as the third party’s motives. The conflicts in Yugoslavia are used to illustrate and test the key implications of the model.