Classical biological control of weeds currently operates under the assumption that biological control agents are safe (i.e., low risk) if they do not directly attack nontarget species. However, recent studies indicate that even highly host-specific biological control agents can impact nontarget species through indirect effects. This finding has profound implications for biological control. To better understand the causes of these interactions and their implications, we evaluate recent case studies of indirect nontarget effects of biological control agents in the context of theoretical work in community ecology. We find that although particular indirect nontarget effects are extremely difficult to predict, all indirect nontarget effects of host specific biological control agents derive from the nature and strength of the interaction between the biological control agent and the pest. Additionally, recent theoretical work suggests that the degree of impact of a biological control agent on nontarget species is proportional to the agent’s abundance, which will be highest for moderately successful control agents. Therefore, the key to safeguarding against indirect nontarget effects of host-specific biological control agents is to ensure the biological control agents are not only host specific, but also efficacious. Biological control agents that greatly reduce their target species while remaining host-specific will reduce their own populations through density-dependent feedbacks that minimize risks to nontarget species.