In one embodiment, a method comprises training at least one model based at least in part on interactions between one or more users and electronic messages sent to addresses associated with the one or more users, receiving a first electronic message sent to a first address associated with a first user, analyzing the first electronic message to generate first feature data, determining one or more characteristics of the first user to generate second feature data, inputting, to the at least one model, the first feature data and the second feature data, and receiving, as output of the at least one model, data indicating whether to output, to the first user, a warning regarding the first electronic message.