The simultaneous handling of 2 tasks requires shielding of the prioritized primary task (T1) from interference caused by the secondary task (T2) processing. Such interactions between tasks (e.g., between-task interference, or crosstalk) depend on the similarity of both tasks and are especially pronounced when both tasks overlap strongly in time. In the present study we investigated whether between-tasks interference can be reduced when specific items do not predict the level of interference but instead the degree of temporal proximity between both tasks. We implemented an item-specific proportion manipulation of temporal task overlap (stimulus onset asynchrony [SOA]). Selected stimuli of T1 predicted high temporal task overlap (short SOAs) in 80% of trials, whereas other stimuli of T1 predicted low temporal task overlap (long SOAs) in 80% of trials. Results showed that the predictive value of T1 stimuli determined the adjustment of T1 shielding. That is, interference from the secondary task was significantly reduced for items predicting high temporal task overlap compared to items predicting low temporal task overlap. It is important to note that task shielding was not initiated by predicting the actual conflict level (i.e., whether T1 and T2 required compatible/incompatible responses) between tasks but by specific items predicting conditions in which 2 tasks are likely to interact (i.e., short vs. long SOA). These findings offer new insights into the specificity of contextual bottom-up regulations of cognitive control.