We provide generalization error bounds for bilevel optimization problems where the inner objective is minimized over a reproducing kernel Hilbert space.