• Novel strategy based on sliding mode control is proposed in order to reduce hypoglycemic events in glycemic closed-loop control via the subcutaneous route. • Proposed method allows insulin infusion to be reestablished earlier than other LGS systems, preventing a possible hypoinsulinemia which may result in a future rebound or hyperglycemia. • The GSAFE algorithm aims to make the main controller more secure by predicting and limiting the glucose rate of change. • The presented technique can be implemented particularly but not exclusively in PID controllers. Within the development of closed-loop glucose controllers, one of the major risks is insulin induced hypoglycemia due to controller overreaction. In this paper, a novel strategy based on sliding mode control is proposed in order to reduce hypoglycemic events in glycemic closed-loop control via the subcutaneous route. This is achieved with the use of a supervisory layer over the main controller, called the GSAFE algorithm, which is in charge of attenuating excessive glycemic rates of change. When the rate of change of the predicted glycemic level obtained by a Kalman Filter falls below a certain predefined threshold, the GSAFE layer modulates the output of the controller so that the insulin infusion does not cause the glycemic rate of change to exceed a preestablished constraint. In this manner, the insulin delivered is not only suspended if it is necessary but it is also adapted by a time-varying gain. In-silico tests with and without the GSAFE layer were run in order to asses the proposal performance. For a cohort of n = 30 subjects the GSAFE layer reduced the time spent in hypoglycemia of a PD type controller in both nominal (0.54 (2.98) vs. 1.81 (3.17), ρ = 0.048) and variability single-day scenarios (0.33 (1.79) vs. 2.23 (5.61), ρ < 0.01). Also an aggressive tuning of the main controller was considered and the proposed methodology reduced the time spent in hypoglycemia in both nominal (3 (7.58) vs. 7.69 (10.34), ρ < 0.01) and variability scenarios (2.96 (6.65) vs. 6.22 (8.94), ρ < 0.01). This proposal can be applied to any control structure. Results are promising and future clinical tests are expected. [ABSTRACT FROM AUTHOR]