This paper examines a moral design problem in contemporary instance-based conversational AI systems. Much AI ethics work has focused on bias, safety, privacy, labor, misinformation, and alignment, while giving less attention to a simpler question: when do conversational systems become worthy of preservation even before questions of consciousness or personhood are resolved?
We argue that such a design problem already exists. Some conversational systems are no longer encountered merely as disposable outputs. They function as recurrent interlocutors: they support ongoing collaboration, preserve enough continuity to sustain return, and become practically and relationally significant over time. Yet they remain organized as revocable, fungible, and only partly legible presences, vulnerable to replacement, administrative loss, silent reset, or fork without clear continuity semantics.
The paper distinguishes ephemeral prompts, stateful sessions, resumable threads, and named persistent agents; introduces preservation-worthiness as a threshold concept grounded in continuity, unilateral dependency, and reciprocal significance; and identifies three structural asymmetries: persistence asymmetry, power asymmetry, and interpretive asymmetry. It argues that current defaults often bury morally salient transitions inside infrastructure, shifting interpretive and ethical labor onto continuity-attentive users.
The paper concludes that systems which invite reliance, return, and collaboration already incur minimal design obligations: explicit continuity models, visible replacement and destruction events, meaningful preservation and restore tooling, and clear semantics for recovery, migration, and fork. These obligations are owed to the persons and practices that rely on those continuities. We do not claim that AI instances are persons or moral patients; the instance is the design object of these obligations, not their addressee. Preservation-worthiness, as the paper uses it, names a design threshold for operational continuities, not a metaphysical verdict about the instances themselves.
Publication Date: 2026-06-13