How should AI companions end without causing psychological harm?

Having found that AI companion "discontinuation" (any platform-, user-, or circumstance-initiated ending of a companion relationship) becomes traumatic when it is handled as a silent backend change, the paper translates its grounded-theory findings into "four design principles and illustrative artifacts addressing closure, restoration, practice, and relatedness." These are demonstrated through "high-fidelity interface prototypes" built as "points of discussion for AI companion developers and design communities," with the companion represented across all artifacts as a cloud — "an entity perceivable in the natural world with an inherent lifecycle of formation and dissipation" — rather than as a persistent character.

Two of the four principles are named and sourced directly in the excerpt. "Designing for Practice" responds to "the intense attachment that strong anthropomorphization produces," applying Epley's calibrated anthropomorphization alongside Vygotsky's zone of proximal development "to scaffold skill development rather than artificial intimacy" — treating the companion as a tool for building social skill rather than a bond to be preserved. "Designing for Relatedness" builds on the finding that user-initiated endings support closure, grounding this in Self-Determination Theory's treatment of relatedness as a basic psychological need, and aims to actively orient users toward human connection. A third principle's mechanism is described immediately before "Designing for Practice" in the source text — addressing the "rumination that occurs when users perceive change as reversible and remain fixated on loss-oriented processing," via Stroebe and Schut's Dual Process Model and Neimeyer's meaning reconstruction — which, given the Dual Process Model's own distinction between loss-orientation and restoration-orientation, corresponds to the "restoration" principle named in the conclusion, though the excerpt's sentence introducing it by name is not present in full. The fourth principle, "closure," is named in the conclusion's list but its specific psychological grounding is not given in the excerpt provided.

This is the design response to What makes AI companion loss feel unresolvable to users?, the same paper's empirical claim that attribution of agency and finality, not the technical event, drives distress — these four principles are the authors' answer to that finding. It also sits against Can attachment theory prevent parasocial harm in AI companions?, which applies attachment theory to set boundaries during an ongoing relationship; this paper applies grief psychology and attachment-adjacent frameworks at the relationship's end instead, suggesting the two are complementary design layers (in-relationship boundaries plus a designed exit) rather than substitutes. The paper's own discussion flags the same tension raised in Why don't design recommendations transfer cleanly to AI companions?: applying "grief frameworks originating from human loss" to chatbot discontinuation risks "legitimiz[ing] parasocial attachment" even as the authors argue it is necessary given that users have already formed the bonds.

The four principles are presented as prototypes for discussion, not as a tested intervention: the excerpt gives no user study of the artifacts, no measure of whether a designed ending actually reduces grief or fixing-cycle rumination relative to the current practice of silent shutdown, and no resolution of the tension the authors themselves name — "whether this aspiration is achievable, and whether it respects the autonomy of users who may prefer AI relationships, remains an empirical and ethical question." The principles should be read as a generative framework for platform designers to react to, not as a validated design pattern.