A method and apparatus for identifying a risky user and a server. The method includes: extracting historical published information of users indicated by preset user identifiers; for each user indicated by each of the user identifiers, performing the following steps of identifying a risky user: extracting a feature vector from the historical published information of the user, and inputting the extracted feature vector to a pre-trained information identifying model to obtain an information identifying result corresponding to the historical published information of the user, the information identifying model being used to characterize a corresponding relation between the feature vector and the information identifying result; and determining the user as a risky user, in response to the information identifying result corresponding to the historical published information of the user indicating the historical published information of the user as risk information.