Systems with redundant control actuators are sometimes arranged so that
a new actuator comes into play if the one normally used becomes saturated.
We call this `daisy-chaining'. Such an arrangement also provides a degree of
fault-tolerance against actuator failures. This talk shows that a
similar property is implicit in constrained predictive control, as it is
usually formulated. Predictive control therefore has a degree of 
implicit fault-tolerance. In order to obtain this property the predictive
control must have explicit constraints on the input levels (actuator 
positions) and the usual disturbance model, which results in integral action
arising in the controller. An attempt will also be made to formalise the 
notions of `redundant actuators' and `actuator reconfiguration'.