ABSTRACT
AI agents are already performing tasks in legal workflows that implicate fiduciary duties: reviewing contracts, analyzing risk, routing compliance decisions, surfacing research. The lawyers and institutions deploying these agents remain fiduciaries to their clients. But when a client or regulator asks ‘prove your AI acted loyally and with care’, there is currently no answer. Not because the duty has disappeared, but because no infrastructure exists to document its fulfillment in a form that survives adversarial scrutiny. This paper proposes the Agent Attestation as the technical operationalization of fiduciary duty for autonomous legal AI agents. The Agent Attestation is a tamper-evident, cryptographically sealed record that captures what an agent did, under what authority it operated, what diligence it exercised, and whether a human reviewed the outcome, structured specifically for legal evidentiary admissibility. The four layers of the Agent Attestation architecture map onto the four core fiduciary duties: the duty of care maps to consensus verification, the duty of loyalty maps to policy binding, the duty of confidentiality maps to the privacy-preserving verification model, and the duty of disclosure maps to the human review record. Crucially, this infrastructure exists today, is deployable at modest cost, and requires no novel technology; only the deliberate application of widely available cryptographic tools to a problem the legal AI community has not yet systematically addressed.
Priore, Kenneth, The Fiduciary’s Footprint: How Legal AI Agents Can Prove They Were Loyal (May 31, 2026).
Leave a Reply