Territorial autonomy is on the rise. More states than ever before are yielding political authority to subnational polities, transforming the fundamental architecture of governance in the modern state. A significant driver of this "devolution revolution" is the perception that territorial autonomy provides a vehicle for peace and democracy in divided societies. Yet, territorial self-governance has proven remarkably deadly in unexpected ways. In contexts as diverse as Nigeria, Russia, Indonesia, and the Democratic Republic of the Congo, devolutions of territorial autonomy have been followed by government military interventions, state-led civilian massacres, and armed conflict. This project investigates the causes and downstream consequences of political violence in the context of territorial autonomy, addressing three interrelated questions: 1) How does territorial autonomy affect state violence against civilians?; 2) How does state violence shape political power and resistance in autonomous territories?; and 3) How and under what conditions does political violence impact the survival of autonomous territories? To answer these questions, I develop a new theory of regional consolidation that explains mass civilian killing as a tool for national actors to reassert authority in autonomous territories. I argue that territorial autonomy activates revanchist imperatives among state leaders, who engage violence to eliminate rival factions from the subnational political arena. While mass civilian killing is an effective tool for consolidating regional authority, it has the unintended consequence of provoking violent political mobilization leading to armed conflict. Armed conflict, in turn, has detrimental consequences for the survival of autonomous regions. When violent conflict breaks out in self-governing territories, it provides a pretense for national actors to implement centralizing reforms that erode autonomy. The dissertation's empirical investigations examine these dynamics through a series of cross-national quantitative analyses. I develop the Territorial Autonomy Dataset, a new geo-referenced dataset that identifies all autonomous regions globally from 1989 to 2019. To overcome methodological challenges related to selection effects and endogeneity, I leverage innovative statistical techniques in the counter-factual estimation framework. The results analyses produce three key findings. First, while the proponents of autonomy imagine stability, the fundamental substance of autonomy reforms creates new incentives for violence against civilians. Second, state violence in autonomous regions nearly always triggers armed uprisings. Finally, violent conflict itself reduces territorial autonomy. Collectively, these findings highlight the circular process through which states counteract power losses in self-governing regions through violence, and how the downstream instability ultimately fuels the structural erosion of autonomous institutions.