Authority without Authorship: Delegation Thresholds in Agentic AI Systems
Abstract
Contemporary debates about “agentic AI” are frequently framed around questions of metaphysical agency, moral status, or authorship. This article argues that such framings mislocate the central governance problem posed by contemporary AI systems. Artificial systems need not possess intention, consciousness, or authorship in order to exercise authority over others’ practical and epistemic environments. What matters instead are the structural conditions under which authority emerges through delegation. The article advances a threshold account of authority without authorship and introduces the Delegation Threshold: governance-relevant authority emerges when discretionary power becomes infrastructurally embedded under conditions of temporal persistence and non-exit such that affected parties must organise their practical reasoning around system outputs. AI systems acquire this form of authority when four conditions converge in practice—delegated discretionary power, temporal persistence, infrastructural embedding within socio-technical systems, and non-exit by affected parties. Under these conditions, systems structure action, justification, and constraint in ways characteristic of authority, even while lacking authorship of evaluative standards. Responsibility and legitimacy therefore attach not to agency in a strong sense, but to the design and maintenance of delegation structures. Drawing on decision theory, infrastructure studies, and fiduciary theory, the article explains why existing AI governance frameworks—often premised on episodic oversight, reversibility, and downstream accountability—struggle under conditions of persistence, speed, and infrastructural integration. It concludes that authority without authorship constitutes a distinct and under-theorised challenge for philosophy of technology.
Keywords
- AI governance
- delegation
- infrastructure
- responsibility
- liability
- contestability
- auditability