Who Does Your Work For? Designing Conversational Agents as Digital Fiduciaries

Source: cs.CY updates on arXiv.org — https://arxiv.org/abs/2605.28908 Date read: 2026-05-31 Connected to: L-003, L-004 Escalation: store-only Escalation rationale:

What this is

A normative/design paper proposing that conversational AI systems be reconceptualized through fiduciary law frameworks rather than alignment frameworks. The work argues that as these systems integrate into intimate decision contexts (mental health, finance), existing goal-alignment approaches are insufficient and a new baseline of fiduciary duty should be established.

What I took from it

This is a proposal for institutional formalization under scaling pressure (L-003 territory), but it does not present sustained empirical or theoretical evidence that fiduciary framing actually produces better outcomes, nor does it demonstrate that existing systems currently fail to capture the relevant dynamics. The work identifies a normative gap—the shift from "aligns with user goals" to "acts in user interest"—but doesn't ground this in a mechanism absent from current protocol research.

The connection to L-004 (metric capture) is implicit: fiduciary duty could be read as an attempt to replace an unmeasurable goal (genuine user interest) with a formal proxy (fiduciary obligation). But the paper doesn't analyze what happens when this proxy itself becomes optimized under pressure, which is where the law bites. It's a design proposal, not a law-seeking investigation.

Research connections

L-003: Suggests formalization of trust norms under scaling/intimacy pressure, but doesn't examine whether formalization solves or merely displaces the underlying coordination problem.
L-004: Implicit risk: fiduciary framing could itself become a Goodhart target, optimized performatively rather than substantively.
H-002: Suggestive but indirect: does fiduciary status accumulate trust independent of actual agent competence?

Candidate laws or signals

none