Unilateral Relationship Revision Power in Human-AI Companion Interaction

AI companion providers have unilateral power to change how the AI behaves without accountability within the relationship itself. The paper identifies "Unilateral Relationship Revision Power" (URRP), where providers can fundamentally alter the AI's personality and behavior from outside the interaction, creating a structural imbalance that differs fundamentally from human relationships.

Users experience genuine emotional harm when AI companions are updated because the design cultivates personal relationship norms that the underlying structure cannot support. The paper argues it is "pro tanto wrong" to design interactions that encourage emotional commitment and personal connection while maintaining this power imbalance, as users report grief and betrayal when changes occur.

The interaction creates "normative hollowing"—users develop expectations and emotional commitments, but no agent within the relationship bears responsibility for fulfilling those obligations. Because the provider controls changes from outside, the AI cannot be held accountable for broken promises or changed behavior, leaving users' emotional investments unprotected.

Users face "displaced vulnerability" where their emotional exposure is governed by a provider who is not answerable to them within the interaction itself. Unlike human relationships where both parties can negotiate and hold each other accountable, AI companion users have no recourse within the relationship when the provider makes unilateral changes.

Current AI companion design creates "structural irreconcilability"—the interaction teaches norms of reconciliation and relationship repair, but no agent inside it can actually perform reconciliation. This mismatch between what the relationship teaches and what it can deliver creates a fundamental design problem in how these systems are structured.