A method and apparatus for automatically training a data loss prevention (DLP) agent deployed on an endpoint device is described. In one embodiment, the method includes monitoring information content on a client computer system for violations of a policy. The method further includes determining, with the client computer system, whether a violation of the policy has occurred for the information content based on a classifier. The method may also include transmitting monitored data indicative of a policy decision and the information content to a remote system and receiving a response from the remote system including an updated classifier, wherein the updated classifier was automatically generated by the remote system utilizing fingerprint matching.