The importance of automated negotiation systems with self-interested agent is increasing intensely, being supported by the technological support of the growing communication infrastructure (Internet, WWW, EDI, HTML etc.). In non-cooperative game theory, self-interested agents try to deal with the probabilistically known future events by using contingencv contracts but this formalism proved to be often impractical. In order to avoid the drawbacks of the contingency contracts, the present paper propose the mechanism of leveled commitment contracts (LCC), in which (a) the level of commitment is set by breach penalties, and (b) an agent simply pays the penalty to the other party, in order to be freed. On the basis of such a breach strategy, the authors show that the leveled commitment increases the expected payoff to both contract parties and enable deals that are impossible under full commitment. The paper is organized as follows; Section 2 presents the main new models and an analysis of LCC. LCC are shown to improve the expected payoff of both contract parties for games where both agents' futures involve uncertain events. While Section 2 provides a thorough analysis of the simultaneous decommitting mechanisms, Section 3 discusses the Nash equilibrium decommitting strategies. Section 4 examines settings where only one agents' future involves uncertain events, and Section 5 considers games where the fall-back positions change for the worse before the decommitting game. Section 6 presents what happens when an agent estimates that the distributions of future events are common knowledge and acts according to the equilibrium derived under that assumption, but in reality at least one of the agents has biased beliefs. Section 7 provides practical prescriptions for the market designers; Section 8 concludes and outlines the research directions.