Detecting AI Agent Use and Abuse