We investigate one possible explanation for observed rates of corrupt behavior namely that individual decision makers who frequently engage in illegal actions may underestimate the overall probability of being caught. This might in particular be true for petty corruption where small amounts of bribes are involved and the detection rate is rather low. To abstract from confounding effects of reciprocal behavior, we design an experiment where a public official decides upon accepting a bribe that leads to a higher present period income while facing the risk of being audited and being left with a considerable lower income in all subsequent periods. Because risk attitudes might differ when putting earned versus endowed income at risk, we compare treatments where participants either receive an endowment beforehand, or earn their income by conducting a real effort task in every period. Independent of the treatments we already find high rates of corruption in very early periods. Risk attitudes measured with a subsequent lottery-choice experiment do not correlate with the behavior observed in the corruption experiment. We explain our findings by a systematic underestimation of the overall probability of being audited. Although detection probability is small in each period, the probability of being caught only once is substantially high when engaging in corrupt behavior on a regular basis. Our findings have important political implications because the underestimation of the total risk involved in engaging in corrupt behavior might nullify measures to fight petty corruption by increased governmental auditing.