Visual short-term memory (vSTM) and cognitive control have largely been studied in isolation, leaving open how vSTM operates in dynamic environments where task priorities shift across feature dimensions. Here, we bridged these domains by embedding change detection (Experiments 1 and 2) and continuous delayed estimation (Experiment 3) paradigms within a task-switching framework in which the relevant feature dimension (colour or orientation) either repeated or switched across trials. Across all experiments, vSTM performance incurred a reliable switch cost on dimension-switch trials relative to repetition trials. Mixture modelling of delayed estimation responses in Experiment 3 revealed that these switch costs did not arise from a reduction in memory precision or an increase in random guessing, but rather from a selective increase in non-target responses reflecting feature-location binding errors. Together, these findings indicate that switching attention between feature dimensions impairs feature-location binding and increases susceptibility to interference from recently attended, currently irrelevant dimensions, demonstrating that vSTM performance is shaped by attentional control dynamics alongside storage capacity limits.