How third-party monitoring meant to fight corruption change the behavior of noncorrupt politicians? To answer this question, I develop a stylized game-theoretic model of political agency in which an incumbent politician decides between implementing a safe policy or an infrastructure project. While the former action immediately produces a payoff for the voter, the latter can substantially improve the future welfare of the voter, but it is also open to corruption and thus subject to an investigation. Despite the investigation is more likely to correctly reveal whether the incumbent engaged in corruption, there is always a positive probability of type-I and -II errors. The voter, being uncertain about whether the incumbent is corrupt, uses the policy choice and safe policy's payoff or investigation outcome to screen out bad politicians. When the incumbent has high reputation, she is re-elected absent bad news and thus the fear of being unjustly perceived as corrupt deters her from implementing the project. In contrast, when her reputation is low, a noncorrupt incumbent has an incentive to implement the project and undergo the investigation to signal her type. Being certain of not being re-elected absent good information, an incumbent may gamble on her future by encouraging voter learning in the aim of highlighting her honesty. However, for this to be an equilibrium, corruption cannot be too profitable as otherwise an honest politician may signal her honesty by not doing the project and thus forgoing large private gains.