This paper demonstrates how corruption negatively impacts policy effectiveness in a program that is explicitly designed to be unsusceptible to bribery, clientelism, and the embezzlement of money: Brazil’s Bolsa Família program. The program’s centralized beneficiary selection process enables me to identify families that were randomly admitted or not admitted to Bolsa Família and to estimate the program’s effectiveness in different years and municipalities. Exploiting a second natural experiment, I then show that Bolsa Família’s effect on school enrollment increases by a third after a municipality has been audited at random. Using a theoretical model, administrative data, and a field experiment with 6,998 registration centers, I find that local corruption increases the probability that families successfully underreport their income when registering for Bolsa Família, making it harder to target the families that benefit most. Ruling out alternative explanations, I show that neither improved school attendance monitoring, administrative processes, infrastructure, complementary programs, tighter governance, nor increased whistleblowing can account for the increase in Bolsa Família’s effectiveness after a municipality has been audited at random. Taken together, these results suggest that income underreporting can explain how, despite the program’s safeguards, local corruption undermines the effectiveness of Bolsa Família.