Artificial General Intelligence and Agency

Should AGI have agency or not? Should it be able to alter its objectives based on the situation it may be dealing with and changing circumstances or should it be entirely goal-driven? Which risk is larger? An interesting overview can be found here.