Governing Relationship State: A Legal Framework for Human-Controlled AI Continuity

Abstract

Persistent human–AI interaction can accumulate corrections, preferences, contextual interpretations, prior decisions, restrictions, authority representations, and other state that shapes later work. That state can remain available even after the factual, legal, or institutional basis for relying on it has changed.

This paper identifies lawful continuing reliance as the central legal object. It proposes proportionate duties of visibility, contestability, provenance, revalidation, restriction, retirement from operative use, transition review, and protection against the reconstitution of previously invalidated state. Those duties would attach principally to actors that control consequential reliance while preserving lawful records, institutional memory, third-party rights, security, intellectual property, and prior lawful use.