Pre-print, Working paper: Successive welfare-maximizing governments must decide whether to implement an affirmative action policy. This affirmative action policy is purported to improve the performance distribution of a targeted group (minority or otherwise), although this improvement becomes trivial when the policy is implemented over a long period of time. Employers pay workers according to their expected productivity, but without (perfectly) observing whether they benefited from affirmative action or not. This has a depressing effect on the wages of non-beneficiaries (whether or not they belong to the targeted group), leading in turn to a feeling of injustice. We find that governments perpetually choose to implement an affirmative action policy, even though the overall feeling of injustice is worse than the purported beneficial effect on the performance of the targeted group (which becomes marginal over time). The explanation we propose is based on a novel moral hazard argument: Each government's actual policy decision is not (perfectly) observed by market participants and thus has no direct effect on wages. Governments therefore do not internalize the depressing effect of an affirmative action policy on wages when choosing to implement it. This outcome is in contrast with a first-best, welfare-maximizing policy plan, in which governments choose to end affirmative action after a certain number of periods, when the purported improvement in the performance distribution of the targeted group becomes small enough.