RNA-protein complexes use diverse binding strategies, ranging from structurally well-defined interfaces to completely disordered regions. Experimental characterization of flexible segments is challenging and can be aided by atomistic molecular dynamics (MD) simulations. Here, we used an extended set of microsecond-scale MD trajectories (400 μs in total) to study two FUS-RNA constructs previously characterized by nuclear magnetic resonance (NMR) spectroscopy. The FUS protein contains a well-structured RNA recognition motif domain followed by a presumably disordered RGG tail that binds RNA stem-loop hairpins. Our simulations not only provide several suggestions complementing the experiments but also reveal major methodological difficulties in studies of such complex RNA-protein interfaces. Despite efforts to stabilize the binding via system-specific force-field adjustments, we have observed progressive distortions of the RNA-protein interface inconsistent with experimental data. We propose that the dynamics is so rich that its converged description is not achievable even upon stabilizing the system. Still, after careful analysis of the trajectories, we have made several suggestions regarding the binding. We identify substates in the RNA loops, which can explain the NMR data. The RGG tail localized in the minor groove remains disordered, sampling countless transient interactions with the RNA. There are long-range couplings among the different elements contributing to the recognition, which can lead to allosteric communication throughout the system. Overall, the RNA-FUS systems form dynamical ensembles that cannot be fully represented by single static structures. Thus, albeit imperfect, MD simulations represent a viable tool to investigate dynamic RNA-protein complexes.