This letter studies an intelligent reflecting surface (IRS)-aided wireless information surveillance system, in which the IRS is leveraged to help the legitimate receiver (LR) intercept the information transmitted from a suspicious transmitter (ST) to a suspicious user (SU). In particular, we consider the challenging scenario where all involved channels are not perfectly known at the LR. We investigate the joint design of the receiver beamformer at the LR and the phase shifts at the IRS for maximization of the worst-case information monitoring rate (WCIMR) by considering the deterministically bounded errors of channel state information (CSI). To solve the resulting non-convex optimization problem, a robust algorithm is proposed based on the general S-procedure, the general sign-definiteness lemma, the penalty-based and successive convex approximation (SCA) techniques. Simulation results demonstrate the effectiveness of the proposed algorithm and also show that the IRS can significantly improve the WCIMR compared to conventional architectures without IRS.