This paper characterizes the symmetric information benchmark for a principal-agent problem under ambiguity, using the smooth ambiguity model of Klibanoff et al. (2005). The problem is formulated as an optimal control problem, and the existence of an optimal wage function is proven. When the principal is risk-averse, the optimal wage is robust to ambiguity: it remains nondecreasing in outcomes regardless of ambiguity preferences or the number of ambiguous states. When the principal is risk-neutral and there are two ambiguous states, robustness still holds if ambiguity affects only one side of the outcome space.