Goal attribution emerges early in the first year of life (Csibra, 2008; Liu & Spelke, 2017; Southgate & Vernetti, 2014). Most research on goal attribution implements experimental scenarios in which the target agent’s goals remain constant. Little is therefore known about whether children update their representations of others’ goals, and specifically, whether children distinguish between goal-directed actions that were halted due to goal interruption and goal abandonment. We tested this using an instrumental helping paradigm. In test trials the experimenter begins to place a toy into a box (initial location). In the interrupted goal-condition, the experimenter encounters an obstacle which interrupts this action. In the abandoned goal-condition, the experimenter states that he would rather place the toy in another box (alternative location). We measure where children help place the toy (initial vs alternative location). Twenty-four children (24-30 months old) were tested using a within-subjects design. Using GLMMs, we found that the odds of helping place the toy in the initial location were over 4 times larger in the interrupted goal-condition compared to the abandoned goal-condition (OR = 4.63, p = .003). This indicates that children differentiate between interrupted and abandoned goals, and this guides their helping in dynamic social contexts.