It is an exciting possibility that dark matter particles are bound states of a strongly interacting dark sector that couples only weakly to the Standard Model. I will present the fundamental requirements on such dark sectors and discuss constraints from cosmology, astrophysics and direct detection experiments. At the LHC, strongly interacting dark sectors give rise to dark showers, which lead to a range of interesting signatures, such as semi-visible jets and displaced vertices. I will discuss how machine learning can be used to improve LHC sensitivity for semi-visible jets and present sensitivity estimates for a monojet search that includes a dynamic graph convolutional neural network as dark shower tagger. Moreover, I will discuss how searches for displaced vertices can be optimised for the case of long-lived dark mesons in the GeV-range.