Method of detecting compromised computers in a network

A computer implemented method of detecting a compromised machine on a network, comprising:

• establishing the network having at least one designated mail server, the mail server relaying all outgoing email messages originating from the network;
• co-locating a compromised machine detection system with the at least one designated mail server to examine the outgoing email messages;
• receiving an email message from a machine on the network;
• classifying the email message as spam or non-spam based on a spam filter or user feedback;
• updating a probability ratio by applying a sequential probability ratio test, wherein the sequential probability test uses the approximate probability that an email message transmitted by a machine on the network is spam if the computer is compromised and the approximate probability that an email message transmitted by the machine is spam if the computer is not compromised, whereby the probability ratio is increased if the email message was classified as spam or decreased if the email message was classified as non-spam, wherein the probability ratio is updated according to the equation
• Λ
• n
• =
• ln
• ⁢
• θ
• 1
• θ
• 0
• ,
• responsive to the email message being classified as spam;
• wherein Λn the probability ratio, θ0 represents the probability that an email message received from a machine on the network is spam given that the machine is not compromised, and θ1 represents the probability that an email message received from a machine on the network is spam given that the machine is compromised;
• indicating the status of the machine as compromised, responsive to the probability ratio being greater than or equal to a first predetermined threshold (B);
• indicating the status of the machine as normal and resetting the probability ratio, responsive to the probability ratio being less than or equal to a second predetermined threshold (A);
• repeating, for a plurality of email messages, the operations of receiving an email message, classifying the email message, updating the probability ratio, and indicating the machine is normal or compromised until the probability ratio is greater than or equal to the first predetermined threshold, wherein the repeated operations are performed on each of the plurality of email messages one at a time as each of the plurality of email messages is received;
• and
• identifying the machine indicated as compromised to a system administrator whereby the compromised machine may be cleaned.