Prolonged exposure to uncontrollable situations can cause individuals to become and remain dysfunctionally passive. This pattern, known as learned helplessness, is associated with depression, making understanding and mitigating it of clinical interest. Learned helplessness has been induced in lab settings using simple tasks, yet real-world control often involves complex, non-linear causal systems where the ability to influence an outcome diverges from the ease with which one can bring about the outcome one wants, and where ascriptions of self-agency are non-trivial, depending on prior mechanistic beliefs and counterfactual inference. We examine learned helplessness in dynamic control environments, allowing participants to interact with causal variables as they influence one another in real time, while we systematically manipulate structure, controllability and reward prevalence. Whilst low levels of practical control reliably induced helpless behaviour, we found this did not depend on reward prevalence or the accuracy of learners' causal models.