Distributed Agency in Human-AI Systems: A Framework for Analyzing Authorship, Control, and Autonomy

This paper proposes a six-layer model (L0–L5) for measuring how authorship and creative control get distributed between humans and AI systems. Current terminology — "AI-generated," "AI-assisted," "human-written" — collapses a complex spectrum into categories that fail creators, studios, unions, and regulators alike. The framework addresses this gap by synthesizing legal requirements (USCO, EU AI Act), professional guild standards (WGA, SAG-AFTRA,DGA), cognitive research on human-AI interaction, and market analysis to produce a shared vocabulary grounded in the rules already governing professional creative work. The model identifies a critical threshold at approximately 90% human agency (L4), where legal, professional, cognitive, and economic standards converge. Above this line, authorship claims are clean and defensible. Below it, they dilute. The paper distinguishes between interaction-level agency (a single prompt-response exchange) and work-level agency (the cumulative creative process), introduces an authorship risk model, and maps how compliance constraints across data protection, talent rights, guild rules, and international regulation segment the market into premium, standard, and budget tiers.