This paper examines how a temporary threat of an anti-corruption audit program targetting local politicians affected spending on public goods and political donations from traditionally corrupt industries. In order to document how politicians respond to a threat of audits, I use an experiment that randomly increased the probability of auditing mayors in Brazil for a year, from 5% to 25% and delivered letters to politicians to make them aware of their status. During the experiment, I find that municipalities with previous experience with the audits program and where mayors had relationships with historically corrupt campaign donors respond to the letters by decreasing their requests for federal transfers directed to construction and non-construction projects (such as purchase of inputs to public services). After the experiment was over, I observe a reversal of the effect for construction projects and persistent negative effects for non-construction projects. Moreover, I find that politicians in the treatment group receive less campaign donations and have lower reelection rates after the experiment. These findings suggest that a temporary threat of punishment of illicit activities alters the political equilibrium and promote persistent changes in the expenditure profile of public goods.