Typically, adults give a primary role to the agent’s intention to harm when performing a moral judgment of accidental harm. By contrast, children often focus on outcomes, underestimating the actor’s mental states when judging someone for his action, and rely on what we suppose to be intuitive and emotional processes. The present study explored the processes involved in the development of the capacity to integrate agents’ intentions into their moral judgment of accidental harm in 5 to 8 year old children. This was done by the use of different metacognitive trainings reinforcing different abilities involved in moral judgments (mentalising abilities, executive abilities or no reinforcement), similar to a paradigm previously used in the field of deductive logic. Children’s moral judgments were gathered before and after the training with non-verbal cartoons depicting agents whose actions differed only based on their causal role or their intention to harm. We demonstrated that a metacognitive training could induce an important shift in children’s moral abilities, showing that only children who were explicitly instructed to not focus too much on the consequences of accidental harm, preferentially weighted the agents’ intentions in their moral judgments. Our findings confirm that children between the ages of 5 and 8 are sensitive to the intention of agents, however, at that age, this ability is insufficient in order to give a mature moral judgment. Our experiment is the first that suggests the critical role of inhibitory resources in processing accidental harm.