Theories of public bureaucracy generally picture the link between politicians and bureaucrats as a top-down relationship of principals and agents, where political authorities try to control their bureaucratic subordinates and minimize agency loss arising from information asymmetries between these actors. Deviating from this standard view, I argue that politicians' dependence on bureaucratic effort for re-election purposes allows bureaucrats to exert political power over their elected principals. If bureaucrats are organized in dense unions with strong tenure protections and deviate in their preferences from their principals, they can collectively shirk effort to influence who their superiors are and what policies they choose. To test this theory, I focus on shirking of US municipal police in response to police budget reforms. I leverage the unprecedented $1 billion cut to NYPD’s budget in the New York City Council for FY2021. Using fine-grained data on NYPD 911 response times together with council members’ preferences on the policy change in difference-in-differences designs, I find that response times increase in districts of non-aligned politicians. Supplementary analyses show that the effect is largely driven by calls where police have more discretion, including alarms, past crimes and vehicle accidents. I further rule out alternative explanations, including changes in the demand for police and police-related protests across NYC. This study informs the theoretical debate on principal-agent relationships in government and sheds light on the importance of political interests to explain policing in US cities.