Excessive Agency refers to the vulnerability arising when Large Language Models (LLMs) are granted more functionality, permissions, or autonomy than necessary, enabling them to perform unintended or harmful actions due to unexpected, ambiguous, or manipulated outputs.
