Fiduciary Duties and AI: Legal Frameworks, Technical Implementation, and Governance

Jun 10, 2026

Two Days at NYU Law: Convening the Fiduciary AI Community

Our team is back from New York, and still digesting two packed days at NYU School of Law. There, on June 4th and 5th, the GliaNet Alliance and the Information Law Institute co-hosted Fiduciary Duties and AI: Legal Frameworks, Technical Implementation, and Governance, with the generous sponsorship of the Future of Life Foundation.

The storied Vanderbilt Hall conference room held a select group of legal scholars, computer scientists, entrepreneurs, civil society organizations, and policy practitioners. All were there for a question that grows more urgent by the month: as AI agents act on our behalf in high-stakes domains like health, finance, education, and consumer life, who owes us loyalty and care, and how do we make those duties real?

This was our second convening at a leading law school. The energy in the room confirmed what we've been seeing all year: the fiduciary AI conversation is no longer a niche. It's a new and promising field taking shape.

The workshop ran under Chatham House rules to allow candid exchange, so we won't attribute views here. But we can share some of the key threads that ran through both days.

When does assistance become abdication? 

A recurring theme was the line between a human fiduciary using AI and a human fiduciary handing off their duty to it. Existing law has long permitted delegation — but the duty itself doesn't transfer. Participants dug into what meaningful supervision actually requires: genuine operational understanding of the tools, the ability to contest and explain outcomes, and human judgment that is more than a rubber stamp. The challenge sharpens as systems grow more autonomous and less legible, even to their builders.

Where should duties attach?

If an AI system gives advice that shapes someone's finances or health, does it remain a mere tool — or has it stepped into a different kind of role? Discussion kept returning to the relational nature of fiduciary obligations: duties run between people and the entities behind the technology, not the software itself. That raises hard questions about complex AI supply chains — developers, deployers, and intermediaries — and which actors are best positioned to prevent harm and answer for it when prevention fails.

Duties you can build and test.

One of the most energizing dimensions of the workshop was the technical one. Sessions examined how fiduciary principles can move from legal theory into running systems: evaluation harnesses that test whether agents actually adhere to duties of loyalty and care, assurance standards for the instructions that govern agent behavior, and so-called “tether” architectures designed so that responsibility can be traced before harm occurs rather than reconstructed afterward. The consistent finding: these duties are becoming measurable — which means they're becoming buildable.

Trust as a business model.

Practitioners in the room made a market case alongside the legal one. Companies are already discovering that verifiable loyalty to customers — no hidden conflicts, no surreptitious monetization of user data — is a durable differentiator in a landscape where trust is scarce and easily destroyed. Certification and trust marks, in the tradition of established consumer-protection signals, surfaced as a practical bridge between principle and market recognition. This is the territory of GliaNet's own Net Fiduciaries™ work, and the conversations affirmed how much appetite exists for it.

Building the movement.

The closing sessions turned to mobilization: how to grow consumer awareness and demand for AI services that genuinely serve the people who use them, and how to ensure accountability lands with the right actors when systems fail. Participants began organizing follow-on collaboration across legal, technical, governance, and industry tracks — early scaffolding for an interdisciplinary community of practice.

That community was the real outcome of these two days. The workshop didn't resolve the open questions — that wasn't the goal. Instead, we established that a serious, cross-disciplinary group is now working on these questions together. More importantly, the fiduciary frame offers something the AI accountability conversation has been missing: a body of law and practice built precisely for relationships of trust, dependence, and asymmetric power.

We're grateful to our co-organizers at NYU Law's Information Law Institute, to the Future of Life Foundation for their sponsorship, and to every participant who brought candor and rigor to the rooms.

Want to be part of this work?

If you're building, researching, funding, or regulating in this space, we'd love to hear from you — reach out to Richard Whitt at richard@glia.net. 

We'll be sharing more from this work in the coming weeks.