Foreign governments frequently intervene in armed conflicts by providing support to rebel organizations against their adversaries. State sponsorship of rebel organizations is less costly than a direct military intervention, but rebels often defy orders, desert fighting or turn guns against their sponsors. Under what conditions do rebels turn against their sponsors? Drawing on principal-agent and organizational theories, I argue that the non-centralized structure of rebel organizations increases the length of the delegation chain from sponsors to rebels, leading to defection. Non-centralized organizations have weak central leadership that is unable to control, monitor or punish its rank and file. Due to this disadvantage, non-centralized rebel organizations are less accountable to their sponsors, cannot credibly commit to rapidly change their policies in response to shifts in the sponsor’s demands and suffer from frequent and destructive quarrels between the top and lower echelons. My argument is tested through the statistical analysis of a novel dataset on Sponsorship of Rebels (SOR), and the case study of Pakistan’s sponsorship of Kashmiri militants, 1989-2004. I find support for my argument that non-centralized organizations are likely to defect against their sponsors. Likewise, the model demonstrates that shared ethnic ties, weak rebels and the existence of transnational support are associated with defection. Finally, the existence of multiple sponsors does not affect the probability of defection.