With a future strong, possibly accelerated, weakening of the Atlantic Meridional Overturning Circulation (AMOC) becoming a plausible scenario, there is a need to develop a reliable warning system for the onset of such an event. While many statistical and physical early warning signals have been suggested for AMOC weakening, there is no warning system yet that demonstrates skill even in a perfect model framework. Here, we present a prototype of a probability based warning system for a strong AMOC decline and illustrate its performance through two examples. In the first, we use a conceptual stochastic AMOC box model where the update rule is equation based. In the second example, we use data from an Earth System Model, where the update rule is based on a machine learning approach. In both examples, a warning probability that the AMOC decreases below a specific threshold within a certain future time horizon is obtained. Using an Observing System Simulation Experiment (OSSE), we show that the warning system has skill compared to persistence as measured by the Brier skill score. This prototype is hence a useful starting point for the development of warning systems that can eventually be useful for policy makers and long-term planning decisions.