The empirical literature on the control of bureaus notes that politicians have difficulty observing bureaucratic output, but this insight is rarely represented informal models. To analyze how bureaus use this uncertainty strategically, we develop a model of expertise-based agenda control, building on the Niskanen (1971) and Miller and Moe (1983) tradition. We show that under some plausible conditions, bureaus will underestimate the benefits, and overestimate the costs, of their programs. In the model, politicians are neither passive nor omniscient: they anticipate the bureau's strategic behavior and establish a monitoring system to counteract it. This possibility of detection changes the bureau's behavior: even imperfect monitoring reduces the bureau's deception of the legislature, whether or not the legislature's demand for the bureau's services is concealed. Moreover, uncertainty by itself matters: if the legislature makes it harder for a risk-averse bureau chief to predict demand or penalty, the bureau will restrain its deception.