Adaptive human behavior builds on prior knowledge about stimulus relevance. Some environments cue such knowledge more than others. To behave adaptively, observers need to flexibly adjust sensory processing to the degree of contextual uncertainty. We hypothesize that the neural basis for these perceptual adjustments consists in the ability of the cortical network to switch back and forth between a rhythmic state that serves selective processing, and a state of elevated asynchronous neural activity that boosts sensitivity. To test this hypothesis, we recorded non-invasive EEG and fMRI BOLD dynamics while 47 healthy young adults performed a parametric visual attention task with varying numbers of relevant stimulus features. Drift-diffusion modeling of response behavior and electrophysiological signatures revealed that greater contextual uncertainty lowered the rate of evidence accumulation while increasing thalamocortical engagement, with concomitant increments in cortical excitability and pupil dilation. As predicted, uncertainty-related processing adjustments were expressed as switches between a state of phase-dependent excitability modulation in the alpha band and a state of increased irregularity of brain dynamics. We conclude that humans dynamically adjust sensory excitability according to the processing fidelity afforded by an upcoming choice, and that neuromodulatory processes involving the thalamus play a key role in adjusting excitability in the human brain.