This paper studies intelligent reflecting surface (IRS) assisted active device detection. Since the locations of the devices are a priori unknown, optimal IRS beam alignment is not possible and a worst-case design for a given coverage area is developed. To this end, we propose a generalized likelihood ratio test (GLRT) detection scheme and an IRS phase-shift design that minimizes the worst-case probability of misdetection. In addition to the proposed optimization-based phase-shift design, we consider two alternative suboptimal designs based on closed-form expressions for the IRS phase shifts. Our performance analysis establishes the superiority of the optimization-based design, especially for large coverage areas. Furthermore, we investigate the impact of scatterers on the proposed line-of-sight based design using simulations.