Perceptual updating has been hypothesised to rely on a network reset modulated by bursts of ascending neuromodulatory neurotransmitters, such as noradrenaline, abruptly altering the brain's susceptibility to changing sensory activity. To test this hypothesis at a large-scale, we analysed an ambiguous figures task using pupillometry and functional magnetic resonance imaging (fMRI). Behaviourally, qualitative shifts in the perceptual interpretation of an ambiguous image were associated with peaks in pupil diameter, an indirect readout of phasic bursts in neuromodulatory tone. We further hypothesised that stimulus ambiguity drives neuromodulatory tone, leading to heightened neural gain, hastening perceptual switches. To explore this hypothesis computationally, we trained a recurrent neural network (RNN) on an analogous perceptual categorisation task, allowing gain to change dynamically with classification uncertainty. As predicted, higher gain accelerated perceptual switching by transiently destabilising the network's dynamical regime in periods of maximal uncertainty. We leveraged a low-dimensional readout of the RNN dynamics to develop two novel macroscale predictions: perceptual switches should occur with peaks in low-dimensional brain state velocity and with a flattened egocentric energy landscape. Using fMRI, we confirmed these predictions, highlighting the role of the neuromodulatory system in the large-scale network reconfigurations mediating adaptive perceptual updates.