Artificial light at night (ALAN) is an anthropogenic pollutant that is intensifying and expanding in marine environments, but experimental studies of community-level effects are generally lacking. The inshore, shallow, and clear-water locations of coral reefs and their diverse photosensitive inhabitants make these ecosystems highly susceptible to biological disturbances; at the same time, their biodiversity and accessibility make them model systems for wider insight. Here, we experimentally manipulated ALAN using underwater LED lights on a Polynesian reef system to investigate the influence on localised nighttime fish communities compared to control sites without ALAN. We collected infrared video censuses of baseline communities prior to manipulation, which we repeated following short-term (mean of three nights) and prolonged (mean of 25 nights) exposures to ALAN. Short-term ALAN exposure did not induce any significant alterations to the nighttime fish community, but prolonged ALAN exposure increased nighttime species richness. Species compositions exposed to prolonged ALAN were more dissimilar from their baseline compared to control sites. The difference between community compositions at prolonged ALAN exposure and control sites was not apparent at the family level; instead, it was observed from the composition of trait guilds. Following prolonged ALAN exposure, more diurnal and nocturnal predatory species (piscivores, invertivores, and planktivores)-particularly those that are site-attached or mobile within reefs-were present in nighttime assemblages. Our experimental findings show that coastal ALAN could cause trophic imbalances and circadian disturbances in localised nighttime reef fish communities. Given that community-wide consequences were only apparent after prolonged ALAN exposure suggests that management of the duration of artificial lighting could potentially be used to reduce impacts on marine ecosystems.