How to identify ecological systems at risk of failure is a central question of modern biology and agriculture. Due to human impacts and global change, there is a growing need for early warning signals that identify when a system is at risk of a state change before changes become irreversible or extremely complex and costly to remediate. Honeybees (Apis mellifera) are an urgent case because our food crops heavily rely on them for pollination and annual bee colony losses reported by beekeepers across the globe are unsustainable. Given enough warning, beekeepers can rescue dying colonies. Here we aim to develop an early warning system for the death of honeybee colonies. We used early warning signals to investigate whether fluctuations and dynamical patterns in internal hive temperature can be used as an early indicator of impending colony failure. Across three distinct datasets, we found that temperature regulation of failing colonies was different enough to distinguish them from healthy colonies before they died. We used a Receiver Operating Characteristic (ROC) approach to compare different predictors and determine the trade-off between sensitivity and specificity in our system. Our study shows that early warning theory can help to identify practical signals of risk of state change even in complex systems that change state relatively rapidly, such as a dying bee colony.