Why an AI agent isn't a common-law agent
The Reporter for the Restatement of Agency explains why an AI agent isn't a common-law agent, and where liability lands instead.
Deborah DeMott, the Reporter for the Restatement (Third) of Agency, wrote a short piece on where agentic AI fits in agency law. Short answer: it doesn't fit cleanly.
Her read: an AI agent is an instrumentality, a tool a developer builds and a user deploys, and it never qualifies as a common-law agent in its own right. Calling it a legal person would only make things worse. But the doctrine isn't stuck. Apparent authority still pins the consequences on whoever held the tool out, and it bites exactly when the tool goes past its instructions. See Moffatt v. Air Canada, where the airline tried to disown its own chatbot.
Her silence is the tell. Liability turns on whether the agent diverged from its instructions, and she never says how anyone proves that.
Research: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6838660