We develop an agency model of organized crime accounting for the main trade-offs involved in the introduction of an accomplice-witness program. We characterize the optimal policy and identify its main determinants in a framework where public officials can be dishonest. Our predictions are tested by using data for Italy before and after the introduction of the 1991 accomplice-witness program. As predicted by the model and the earlier antitrust literature, the program appears to have strengthened deterrence and enhanced prosecution. Moreover, consistent with a novel prediction of our theory, the evidence suggests that the program efficacy is affected by the judicial system efficiency.