A system and method are disclosed herein for providing mitigation of fraud in a hosted messaging service while having minimal impact on authorized messaging users. The method includes a system for detecting potential fraud based on multiple and configurable fraud indicators as well as historical data, which can be customized for individual users or groups. The system can terminate in-process messages that are potentially fraudulent and reset the network access credentials for the affected user accounts or devices that have been potentially compromised. The system uses historical data to block further messages from the compromised user accounts or devices to specific destination addresses where the presumed fraudulent messaging activity was directed. In a further aspect, the system and method can automatically reset the network access credentials for authorized users with minimal downtime.