Under which conditions do anti-corruption policies effectively reduce bureau-cratic corruption? Previous studies find that anti-corruption audits are effective in disciplining politicians, but their impact on bureaucrats is unclear. We leverage 10 years of randomized audits and the careers of 275 thousand Brazilian municipal of-ficials. We find that even when strong evidence of corruption is found, audits do not have observable implications for bureaucratic careers, such as dismissals or depar-tures. To investigate whether audits trigger unobservable reductions in corruption or have long-term disciplining effects, we propose a model of corruption with career concerns that we estimate structurally. We rule out that audits have unobservable reductions in corruption, and our results are consistent with either large disciplin-ing effects, or limited effectiveness. We identify strong complementarities among the program’s components, suggesting that multi-pronged approaches combining increases in the frequency of audits with tougher sanctions are most effective at reducing corruption.