A computer-implemented method for deterring scams including: monitoring by a cognitive engine agent incoming messages for scam messages; receiving by a messaging server an incoming message having a sender of the incoming message; identifying by the cognitive engine agent the incoming message as a scam message; and replying by the cognitive engine agent in cooperation with the message server to the scam message by initiating a message conversation with the sender of the scam message, the message conversation including one or more reply messages to the sender of the scam message replying to the scam message and any subsequent scam messages from the sender of the scam message with each reply message being an intelligent, interactive message using natural unscripted language to appear as if the one or more reply messages was written by a human and is responsive to a content of a scam embodied in the scam message.