Catastrophic failure is a critically-important phenomenon in the brittle Earth on a variety of scales, from human-induced seismicity to natural landslides, volcanic eruptions and earthquakes.It is invariably associated with the structural concentration of damage in the form of smaller faults and fractures on localised zones of deformation, eventually resulting in system-sized brittle failure along a distinct and emergent fault plane. However, the process of localisation is not well understood – smaller cracks spontaneously self-organise along the incipient fault plane, often immediately before failure, but the precise mechanisms involved have yet to be determined. Many questions remain, including: Q1 How do cracks, pores and grain boundaries interact locally with the applied stress field to cause catastrophic failure to occur at a specific place, orientation and time?; Q2 what dictates the relative importance of quasi-static and dynamic processes?; and Q3 why can we detect precursors to catastrophic failure only in some cases? Here we will address these questions directly by (a) imaging the localisation process at high resolution, using a newly-developed x-ray transparent deformation cell and fast synchrotron x-ray micro-tomography and (b) applying state of the art acoustic monitoring and source location methods. We will deliberately slow the process to better resolve its temporal evolution, and to investigate the strain-rate dependence of the underlying mechanisms, using rapid electronic monitoring and feedback control. This will provide unprecedented direct observation of the relevant mechanisms, including the contribution of seismic (local cracking producing acoustic emissions) and aseismic (elastic loading and silent irreversible damage) processes. This innovative combination of techniques is timely, feasible, and is likely to transform our understanding of microscopic processes and their control of system-size failure. The results will provide interpretive models for similar processes in natural and human-induced seismicity, including scale-model tests of strategies for managing the risk of large induced events.