Designing Fiduciary Artificial Intelligence

Fiduciary duty provides a legal framework for AI accountability: The paper argues that organizations using AI should be bound by fiduciary duties—legal obligations to act with loyalty and care toward users—similar to how financial advisors must act in clients' best interests, rather than relying solely on user consent which may be incomplete or inadequate for complex technical systems.

A four-step design procedure is proposed: Designers of Fiduciary AI should (1) understand the system's context, (2) identify who the principals (users/stakeholders) are, (3) assess what serves those principals' best interests, and (4) ensure both loyalty to those interests and appropriate care in implementation—providing a concrete methodology for building trustworthy AI systems.

Fiduciary AI addresses consent limitations: The paper identifies a critical gap in current AI governance: data subject consent alone is insufficient for protecting users interacting with complex AI systems, and fiduciary duties can fill this gap by establishing legal obligations that go beyond what users can reasonably understand or agree to.

Connection to existing trustworthy AI principles: The fiduciary framework aligns with established dimensions of Trustworthy AI including privacy and alignment, suggesting that fiduciary duties can serve as a practical bridge between abstract AI ethics principles and concrete legal and technical requirements.