Drowsiness alters the neural dynamics but not the core computations of multisensory integration

Fluctuations in alertness shape perception and behaviour, yet how they affect the brain’s ability to integrate information across senses remains poorly understood. Here we investigated whether multisensory integration is preserved as humans transition from wakefulness to sleep. Participants (18 females, 8 males) performed an audio–tactile detection task while electroencephalography tracked spontaneous declines in alertness. Behaviourally, multisensory stimulation continued to facilitate responses during drowsiness, despite slowing and increased omissions. Although race-model predictions were consistent with preserved evidence for multisensory integration, event-related signatures of multisensory interactions were attenuated or no longer detectable during drowsiness. Crucially, multivariate decoding showed that multisensory neural representations remained detectable, although less temporally stable, and early cross-state generalization revealed shared representational structure across alertness levels, indicating preserved core multisensory computations. These findings show that multisensory integration remains functional but dynamically altered as alertness declines, revealing how the brain maintains adaptive behaviour during the wake-to-sleep transition. Significance Statement Multisensory integration has often been described as a stable and automatic perceptual process. However, growing evidence indicates that it can be modulated by ongoing brain states and cognitive demands. Here, we ask whether this computation remains invariant as alertness declines. We show that multisensory integration is neither abolished nor fully invariant during drowsiness. Instead, behavioural benefits are maintained despite less stable neural dynamics. Although event-related responses are attenuated, core multisensory representations remain shared across levels of alertness. These findings indicate that multisensory integration is preserved across fluctuations in alertness, while their neural dynamics are flexibly altered as alertness decreases.