Agent security through software mutation

Enhancements to electric power distribution grid reliability and recovery in the face of disruptions are being made possible through the incorporation of agent-based control systems. As more agents replace human controllers, the potential for catastrophic consequences of a cyber-attack are increased. One means of preventing, detecting and recovering from such incidents uses program mutation in a novel way to form the basis of a security framework. The Mutation described in this paper modifies the memory footprint of an executing process without altering its runtime behavior. Mutated processes are shown to be resistant to attacks resulting in arbitrary code execution. Instead of Byzantine failures, attacks on mutated processes result in the more easily detectable crash failure. Replication of mutated code is used with a majority voting algorithm to prevent attack consequences from occurring when an attack eventually bypasses the prevention mechanisms. An example is provided to demonstrate the code modfications done by our mutation engine. The example also shows how the mutations are used against buffer overflow attacks which are generally considered the most common form of attack. Agent controlled power distribution grids require protection from intrusion. Whether the attack originates from a careless employee of from a cyber terrorist, the consequences could be catastrophic. This method can be used to combat known and future forms of attack by giving the system the ability to protect control agents, detect attacks against them, and recover from attacks before serious consequences are realized.