Across many studies, researchers have found that representations in working memory (WM) can guide visual attention toward items that match the features of the WM contents. While some researchers have contended that this occurs involuntarily, others have suggested that the impact of WM contents on attention can be strategically controlled. Here, we varied the probability that WM items would coincide with either targets or distractors in a visual search task to examine (1) whether participants could intentionally enhance or inhibit the influence of WM items on attention and (2) whether cognitive control over WM biases would also affect access to the memory contents in a surprise recognition test. We found visual search to be faster when the WM item coincided with the search target, and this effect was enhanced when the memory item reliably predicted the location of the target. Conversely, visual search was slowed when the memory item coincided with a search distractor, and this effect was diminished, but not abolished, when the memory item was reliably associated with distractors. This strategic dampening of the influence of WM items on attention came at a price to memory, however, as participants were slowest to perform WM recognition tests on blocks in which the WM contents were consistently invalid. These results document that attentional capture by WM contents is partly, but not fully, malleable by top-down control, which appears to adjust the state of the WM contents to optimize search behavior. These data illustrate the role of cognitive control in modulating the strength of WM biases of selection, and they support a tight coupling between WM and attention.