One of the defining characteristics of much contemporary nonconsequentialist thought is skepticism about interpersonal aggregation. Aggregative moral reasoning frequently yields counterintuitive implications, especially in cases where it enjoins us to let a few people suffer large losses so that many people can enjoy smaller benefits. To avoid these implications, leading nonconsequentialists have put forward “competing claims” models of moral rightness, according to which morality requires us to determine, by a series of pairwise comparisons, that action or policy which satisfies the strongest individual claim or generates the weakest individual complaint. The fullest development of this anti-aggregative approach is the contractualist moral theory defended by T.M. Scanlon. My paper examines a problem for the competing claims model: Outside the context of crime and war, most instances of serious harm don’t occur in situations where someone trades off certain harms to known individuals against certain benefits to other known individuals. Rather, most harm to others results from risks that some agent imposed on them in pursuit of some benefit for them, for the agent, or for others. The question is whether contractualists can provide a plausible account of such risky actions while honoring their anti-aggregative commitments. I focus on one class of risky actions – what I call instances of social risk – which are pervasive, indeed ineliminable from the conduct of large-scale public policy, yet prima facie troubling for the contractualist. They are characterized by the following four features: (1) The risky action will affect a large number of individuals. Because of this, it is virtually certain that some people will end up being burdened by it. (2) The individual losses to those who are burdened are considerably greater than the individual gains for those who are benefited. (3) The action-type in question is rare, or rarely affects the same people twice; as a result, we cannot assume that over time almost everyone will benefit from a principle that permits actions of this type to be performed. (4) The risky action is intuitively permissible. I submit that many actions in public policy, particularly in the domains of public health and of risk management, share these four features. The problem of social risk forces the contractualist to confront a question about his theory that, I argue, has not received a fully satisfactory answer to date: Are an individual’s personal reasons for rejecting a principle permitting the imposition of a given social risk a function of the prospect that the risky action gives to each person ex ante, allowing us to discount both benefits and harms by their improbability of occurring? Or is justifiability to each person a function of the action’s outcomes ex post? In that case, what would matter morally isn’t the ex ante likelihood of any given individual’s being benefited or harmed by the action, but rather the near certainty that, ex post, some persons will turn out to have been harmed by the action while others are benefited. The contractualist faces an apparent dilemma: if he rejects the ex ante view of justification, as Scanlon explicitly does in What We Owe to Each Other, and appeals instead to an ex post view, he seems committed to the intuitively unappealing conclusion that most instances of social risk are unjustifiable. On the other hand, if he embraces an ex ante view, Scanlon fears that this will move his contractualist theory too close to those aggregative views that he wished to escape from in the first place. In my paper, I elaborate a version of contractualism, what I call stage-wise ex ante contractualism, that solves the problem of social risk without resorting to interpersonal aggregation. My solution, however, comes at a price: Contractualism can no longer claim to be the correct account of what it is for an action to be morally wrong, all things considered. Instead, contractualism can at most claim to capture one class of moral considerations that contribute to making actions right or wrong all things considered.