AI agents miss agency law's loyalty duties, paper argues
TL;DR
- The paper maps implied authority in agency law onto AI value-alignment, calling both responses to the problem of imperfect objective specification.
- Authors Riedl and Desai argue computer science has under-theorized loyalty and third-party duties that are central to the law of agency.
- They warn the gap could cause unintended effects in AI e-commerce agents, and propose a path for responsible development and deployment.
Computer science and law have both worked out how to make an agent do what its principal wants and no more, but only one has thought through what happens when that agent transacts with strangers. In a new arXiv preprint, Mark O. Riedl and Deven R. Desai argue the AI research community has borrowed the first half of agency doctrine and left the second untouched.
The authors frame the arrival of agentic AI as a technical and socio-legal problem at once, lining up implied authority in agency law with the AI principle of value-alignment and treating both as responses to imperfect objective specification. On that half of the comparison, they see the two fields in rough sync.
The gap they name is loyalty and disclosure. The paper states that "computer science has under-theorized issues related to questions of loyalty and to third parties that interact with an agent, both of which are central parts of the law of agency." That matters most, they argue, in AI ecommerce agents, where ignoring those duties "can result in unintended effects."
The piece is a conceptual mapping, not an empirical study: no companies named, no benchmarks published, and the path forward it proposes is laid out rather than tested. It is scheduled for the 2025 AAAI Conference on AI, Ethics, and Society.
Shared on Bluesky by 1 AI expert
Originally reported by arxiv.org
Read the original article →Original headline: AI Agents and the Law