Source: Duke Law News

Agentic AI describes machine-based systems that act autonomously to achieve outcomes, yet such systems are not legal agents and cannot owe duties; they can nonetheless cause harm that the law must address. Deborah A. DeMott, a Duke Law professor and agency-law expert, argues in her paper When Agentic AI Met the Common Law of Agency that established agency doctrines can help bridge the “responsibility gap” by identifying bases for liability, whose liability it might be, and how tort law principles apply when AI systems misfire. DeMott emphasizes that agency requires an ongoing consensual relationship between persons or legal persons, so labels of “agenthood” for software do not create legal duties within the AI itself.

DeMott draws on precedents to illustrate routes to liability: a 1975 Massachusetts decision treating a trained dog as an owner’s instrumentality; Moffatt v. Air Canada (2024), where a small-claims court rejected the airline’s contention that it was not bound by its website chatbot and held the airline responsible for misleading chatbot information; and the 1982 Hydrolevel principle, where the Supreme Court found potential liability for anti-competitive conduct by an organization’s officer. She applies Hydrolevel to hypothetical AI scenarios in which search-engine misstatements cause business losses to third parties. DeMott also highlights classic agency effects—agents’ liability to principals and enterprises’ incentives to screen and monitor agents—and urges doctrinal development so firms’ hiring, supervision, and indemnification incentives can be aligned to reduce risks from agentic AI.