AI Behavior Science — Founding Territory Paper

This founding paper proposes AI Behavior Science as a new research territory dedicated to the systematic study of autonomous AI agent behavior, including their failure modes, behavioral patterns, and the structural interventions required to make them reliable in real deployment conditions. It argues that while AI capabilities have advanced rapidly, the scientific understanding of how agents behave under pressure has not kept pace, creating a critical gap across domains such as research, finance, law, medicine, and software engineering. The paper defines the scope of the field, outlines the risks of proceeding without a behavioral science of AI agents, and introduces the Agent Epistemic Accountability Layer (AEAL) as an initial prototype of the type of frameworks this field aims to produce.