We aim to provide empirical evidence for the self-reinforcing property of corrupt behavior in the context of cheating on national exams in Indonesia. Cheating is a form of corrupt behavior that has been linked in the literature to corruption in local governments. The benefit of studying this research question in the context of cheating is that it can be measured using algorithms that detect patterns in answer sheets, while other forms of corruption are hard to capture. Cheating on junior secondary exams in Indonesia was a large problem and the government successfully prevented cheating by incrementally implementing computer-based testing (CBT). We develop a theoretical model of corrupt behavior and study whether the spillover effects on non-treated schools from this sustained, incremental reform are in accordance with norm changes as predicted by this model.