Persistent Memory Defaults and Hybrid Conversation-Level Overrides: Extending User-Controlled Memory Governance in Generative Artificial Intelligence

The development of generative artificial intelligence (AI) has increasingly incorporated persistent memory, cross-conversation personalization, and conversational history retrieval to support continuity between users and intelligent systems. Although these capabilities improve contextual relevance and reduce repetitive information exchange, they also create a fundamental memory-governance challenge: users may need different memory configurations for different conversations, while repeatedly configuring those settings can become inconvenient and inconsistent. A conversation may require access to established personal memories, another may require an independent context, and a third may require access to only selected categories of historical information. A purely conversation-specific configuration provides flexibility but does not automatically establish a reusable default for subsequent interactions. Conversely, a single persistent account-level setting may provide consistency but insufficient flexibility when users move between different tasks and domains. Building directly upon the previously published framework Memory-Isolated, History-Preserved, and Future-Integrable Conversations: A User-Controlled Memory Architecture for Generative Artificial Intelligence (MIHFI), this paper proposes two complementary memory-management modes: Persistent Default for Subsequent Conversations (Mode B) and Hybrid Configuration (Mode C). The earlier MIHFI framework established the conceptual separation of historical memory retrieval, conversational history preservation, and future memory-source eligibility, providing a foundation for independent memory control within an individual conversation. The present study extends that foundation from conversation-level configuration to cross-conversation policy inheritance and controlled local overrides. Mode B enables users to define a reusable memory configuration that is automatically inherited by newly created conversations until the persistent default is modified. Mode C extends this mechanism by allowing an individual conversation to temporarily override selected elements of the inherited configuration without modifying the persistent default or affecting other conversations. The proposed architecture introduces reusable memory profiles, versioned policy inheritance, field-level override precedence, and explicit separation between global defaults and local conversational decisions. It also distinguishes memory-reading permissions from conversation preservation and future memory integration, thereby maintaining compatibility with the three-dimensional MIHFI model. A formal system model, architectural principles, operational workflows, research propositions, and an experimentally testable evaluation methodology are presented. Potential evaluation dimensions include inheritance accuracy, override independence, unauthorized retrieval, configuration consistency, user autonomy, privacy perception, interaction efficiency, and cognitive load. This paper does not claim an implemented system or experimentally demonstrated performance improvements. Rather, it proposes a conceptual architecture for making persistent conversational memory configurable, reusable, and locally adaptable. Its central proposition is that users should be able to establish how future conversations normally access memory while retaining the ability to independently redefine the memory context of any particular conversation.