Explosive volcanic eruptions can eject large amounts of ash into the atmosphere, posing a serious threat to populations living near the volcano. The abrupt occurrence of such events requires a rapid response and proper volcanic hazard evaluation. Current monitoring procedures still require human intervention, which often results in significant delays between the occurrence of an eruption and notifications being dispatched. We show how dedicated infrasound array processing can be used to detect and notify the authorities, automatically and in real time, of the onset of explosive eruptions. Conceptually, our method relies on the strong coupling between infrasound and the explosive process, and it is not based on probabilistic considerations but on the ability infrasound has to detect the early stage of the explosive phase. This procedure has been tested for the last 8 years, and it is currently applied to issue early warnings for explosive eruptions at Etna Volcano. We show that the system is able to provide a prealert ~1 hr before the eruption, and it has a 96.6% success rate, with only 1.7% false positive alerts and no false negative alerts. This is, to our knowledge, the first example of an operational early warning system totally based on an unsupervised algorithm that provides automatic notifications of eruptions to a government agency. We show that the same early warning concept might be applicable to arrays at large distances (>500 km), suggesting that infrasound could be successfully used to issue automatic notifications of ongoing eruptions at regional to global scales.