We argue that simple delegation rules are optimal in the presence of uncertainty in how agents manipulate delegation rules for their own private interest by considering a max-min delegation design problem, in which the principal has limited information about the agent’s preferences. In particular, the principal only has limited information on the agent’s preferred actions in each state but does not know how the agent trades off among suboptimal options once constrained. The principal has a prior over state and evaluates a delegation set by its worst expected payoff among all possible preferences of the agent. We show that the optimal delegation set is always convex, which implies that interval delegation is always optimal when the action space is in one dimension. In high dimension, the uncertainty of manipulation can be so large that the principal may delegate actions that are never optimal in any state or even delegate the entire action sets. Joint with Ricardo Alonso and Ju Hu