Within the human in the loop context, the realization of a task will not only be the accomplishment of human operator or the autonomous agent acting on his behalf, but rather of both entities, and in which they will have the same possibilities to propose, suspend, refuse, and stop each other. However, this cohabitation is both rich and complex, owing to the fact that the human and the agent are bound to not only agree on the various levels of realization of the task inside the same loop, but also to manage the autonomy -who controls who-. Hence, this gives rise to a dilemma between the autonomy of an agent that is useful but risky and the fallibility of human in control of the decision-making. The issue then is to work out a computational model of an agent authority sharing, for the purpose of dynamically and safely transferring decision-making control to the human user