Cognitive control processes can mirror fast and dynamic adaptations towards a changing environment. When performing dual tasks (DT), control states are established that help to manage between-task interference. One frequently investigated aspect of cognitive control concerns the generality and persistence of these control states. Such induced processing adjustments can persist even if new stimuli or task sets are introduced (Surrey et al., 2017). In the present study, we investigated to which extent control states are maintained even if major characteristics of the task context change, e.g., if one of the tasks of a dual task becomes irrelevant. For this, we adapted the fade-out paradigm (Mayr & Liebscher, 2001) to a dual-task setting. In a first experiment, participants (N = 40) underwent blocks of DT to establish dual-task-specific control states. In the following block, one of the tasks was eliminated. Results showed that performance in this fade-out block did not immediately drop to single-task performance (fade-out costs) representing the persistence of control states. In a second experiment (N = 80), the proportion of between-task interference in DT blocks was manipulated (75% vs. 25% incongruent trials) between participants to establish conflict-biased control states. These conflict-induced control states did not modulate fade-out costs. Nevertheless, the control adaptations persisted because the size of between-task interference during fade-out depended on the conflict proportion manipulation. These findings suggest that control states influence multitasking performance at two distinct levels – the task-set level and the level of between-task interference. Implementations of this new evidence are discussed.