In type 1 diabetes (T1D), insulin replacement therapy should ideally replicate endogenous insulin secretion, but achieving this goal requires frequent adjustments to insulin delivery based on glucose levels and trends, carbohydrate intake and physical activity. An overriding concern for people taking insulin is hypoglycaemia, which remains the most feared consequence of insulin therapy and limits therapy intensification options. Although fully automated systems that achieve consistent euglycaemia in T1D remain an elusive goal, improvements in continuous glucose monitoring (CGM) sensors and control algorithms have enabled semi-automated systems that lower the risk of hypoglycaemia, especially nocturnal hypoglycaemia. The present review focuses on an important advance in insulin delivery systems: the use of CGM data to stop insulin delivery in the presence of hypoglycaemia. Although conceptually simple, this strategy represents a critical step in the journey toward a fully closed-loop artificial pancreas; the next steps in this journey are also discussed.