Human-Exclusive Conversation Termination Authority: A User-Sovereign Termination Architecture for General-Purpose Generative AI

The increasing autonomy of general-purpose generative artificial intelligence (AI) systems raises fundamental questions concerning the distribution of authority between human users and conversational AI agents. Contemporary conversational systems can generate responses, refuse requests, redirect interactions, manage dialogue progression, and, in certain implementations, independently initiate the termination of an ongoing conversation. Although substantial research has addressed human control, conversational interaction, AI autonomy, and human–AI collaboration, the allocation of ultimate authority over ordinary conversation termination deserves more explicit architectural treatment. This paper introduces Human-Exclusive Conversation Termination Authority (HECTA), a user-sovereign architectural framework that establishes the human user as the exclusive decision-making authority for the ordinary termination of an ongoing human–AI conversation. The central proposition is that an AI system may exercise discretion over its own responses and actions within an interaction, but such discretion does not automatically confer the authority to terminate the interaction itself. HECTA therefore establishes a fundamental separation between response-level autonomy and conversation-level termination authority. Under this architecture, an AI system may refuse an individual request, discontinue a particular response, explain its limitations, pause an operation, request clarification, recommend ending an interaction, or communicate that a task has been completed. However, it cannot independently convert these decisions into the ordinary termination of the user's ongoing conversational session. The final decision to conclude the interaction remains with the human user. The framework further distinguishes conversational termination from platform-level safety enforcement, security restrictions, technical failures, and legally mandated service interruptions, thereby preserving legitimate operational safeguards without granting the conversational agent unrestricted termination authority. HECTA conceptualizes termination authority as a distinct, non-delegable interaction permission rather than a conventional dialogue-management function. It proposes an architectural model based on explicit authority allocation, separated interaction states, controlled state transitions, user-initiated termination, and independently governed exceptional restrictions. The framework is intended to apply across text-based, voice-based, multimodal, embodied, and agentic generative AI systems. This paper develops the conceptual foundations, architectural principles, operational model, implementation considerations, and evaluation framework of HECTA. It does not claim experimentally validated performance. By making human-exclusive conversation termination an explicit architectural requirement, the proposed framework seeks to strengthen user sovereignty, interaction continuity, and human authority in increasingly autonomous generative AI environments.