When completing a task for a prolonged period, our ability to sustain attention fluctuates over time. In mice, errors and biased responses cluster in time rather than occurring randomly, indicating that animals transition into and out of disengaged states over the course of an experimental session. What neural and physiological processes trigger these state transitions? Here, we investigated the role of pupil-linked arousal, as relatively high or low levels of arousal are associated with poor task performance, and there is evidence that the activity of ascending arousal systems triggers largescale cortical network changes. We used a public dataset of 140 mice performing a perceptual decision-making task, including extracellular neural recordings alongside behavioural and pupil responses. We applied hidden Markov models to whole-session time courses of response times. This provided trial-by-trial assignments to an engaged state with fast response times or a disengaged state with longer and more variable response times. Preliminary results show that disengaged trials are associated with larger and more variable baseline pupil, and suggest that activity of arousal systems causes increases in pupil size prior to state transitions. These findings will provide a starting point for exploring the cortical, subcortical and neuromodulatory processes preceding task (dis)engagement, and discovering early warning signals for transitioning into disengaged task performance.