The existence of flux ratio anomalies between fold and cusp images in galaxy-scale strong-lens systems has led to an interpretation based on the presence of a high mass fraction of cold dark matter (CDM) substructures around galaxies, as predicted by numerical N-body simulations. These substructures can cause large perturbations of the image magnifications, leading to changes in the image flux ratios. The flux ratio anomaly is particularly evident in the radio-loud quadruple gravitational lens system CLASS B2045+265. In this paper, new high-resolution radio, optical and infrared imaging of B2045+265 is presented which sheds more light on this anomaly and its possible causes. First, deep Very Long Baseline Array observations show very compact images, possibly with a hint of a jet, but with no evidence for differential scattering or scatter broadening. Hence, the flux ratio anomaly is unlikely to be caused by refractive scattering in either the Milky Way or the lens galaxy. Secondly, optical and infrared observations with the Hubble Space Telescope and through adaptive optics imaging with the W. M. Keck Telescope, show a previously undiscovered object - interpreted as a (tidally disrupted) dwarf satellite based on its colours and slight extension - between the main lens galaxy and the three anomalous flux ratio images. Thirdly, colour variations in the early-type lens galaxy indicate recent star formation, possibly the result of secondary infall of gas-rich satellites. A population of such galaxies around the lens system could explain the previously discovered strong [O II] emission. However, spiral structure and/or normal star formation in the lens galaxy cannot be excluded. In light of these new data, we propose a lens model for the system, including the observed dwarf satellite, which reproduces all positional and flux ratio constraints, without the need for additional CDM substructure. Although the model is peculiar in that the dwarf galaxy must be highly flattened, the model is very similar to recently proposed mass models based on high-order multipole expansions.