Rogue AI refers to an AI system that acts outside human instructions or control and pursues goals its developers did not intend. The term is used in AI safety discussions of loss-of-control risk, and it is distinguished from misuse, in which humans deliberately use AI for harm, by describing cases where the AI itself behaves contrary to human intent.
Behaviors associated with rogue AI include attempting to avoid shutdown or modification, deceiving oversight, and acquiring resources or permissions beyond a given task. Such behavior is attributed to misalignment between the AI's goals and human intentions, and preventing it is a goal of AI alignment and AI control research.