We consider a principal-agent model with moral hazard, bilateral risk-neutrality, and limited liability. The principal knows only some of the actions the agent can take and evaluates contracts by their guaranteed payoff over possible unknown actions. We show that linear contracts are a robustly optimal way to incentivize the agent: any randomization over contracts can be improved by making each contract in its support linear. We then identify an optimal random linear contract characterized by a single parameter that bounds its continuous support. Several corollaries arise: the gain from randomization can be arbitrarily large; optimal randomization does not require commitment; and screening cannot improve the principal’s guarantee.