Answers are grounded in this paper and cite their sources.
Generative artificial intelligence (GenAI) systems are increasingly evolving from isolated question-answering tools into persistent conversational environments capable of retaining user preferences, referencing previous conversations, maintaining personalized memory, and integrating historical contextual information into newly initiated interactions. Although these capabilities improve continuity and personalization, they introduce an important security concern: creating a new conversation does not necessarily mean starting with an independent or information-isolated context. A newly initiated conversation may receive information originating from previous conversations through persistent memory, chat-history references, user profiles, contextual summaries, custom instructions, or other cross-conversation information mechanisms. Consequently, an individual who gains temporary access to an already authenticated AI application may potentially initiate a new conversation and obtain information derived from the account holder's previous interactions, even without directly opening the original conversations. Existing application-level authentication and conversation-level access restrictions do not necessarily address this specific security boundary. This paper proposes Per-Conversation Biometric Authorization for Generative Artificial Intelligence (PCBA-GAI), a novel framework that integrates independent biometric authorization into both new conversation creation and access to historical conversational information. Under the proposed framework, every activation of the ordinary new-conversation function requires a separate biometric verification event, using supported authentication modalities such as fingerprint or facial recognition. More importantly, successful authorization for creating a new conversation does not automatically authorize unrestricted access to the user's historical memories or cross-conversation contextual information. The framework introduces a second authorization boundary governing whether a newly created conversation may retrieve, reference, or utilize information originating from previous conversations. Each conversation receives an independent authorization context, with explicit policies governing historical memory access, sensitive information retrieval, contextual inheritance, and authorization revocation. The framework further supports model-independent authorization, ensuring that conversations initiated through different generative AI models cannot inherit authorization from previously authenticated interactions. To the best of our knowledge, this paper presents the first explicitly formulated research framework integrating mandatory per-conversation biometric authorization with independently controlled cross-conversation memory access in generative AI environments. By addressing both unauthorized conversation initiation and unauthorized historical-context exposure, PCBA-GAI establishes a foundation for more granular privacy and identity governance in persistent, personalized, and multi-model artificial intelligence systems.
No takes yet. Share an insight, caveat, or question.
Jincheng Zhang (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: