When AI Speaks in a Person's Voice: Contextual Identifiability and Technical Control in Chinese Law

Abstract

Voice cloning technology converts a situated act of speaking into a reusable capacity to make a person appear to speak in settings that person did not choose, raising profound questions about identity, autonomy, and legal protection in the age of generative artificial intelligence. Existing rules centred on copied recordings capture only part of that wrong, leaving significant gaps in how the law addresses synthetic speech. Drawing on Article 1023(2) of the Chinese Civil Code and the Beijing Internet Court's 2024 Yin Mouzhen judgment, this article develops a sequential account of synthetic voice as relational digital identity. The protected interest is identified as personality autonomy: a bounded authority to determine whether, by whom, and in which contexts an identity-bearing voice may represent the person, rather than ownership of accent, timbre, or style. Contextual identifiability supplies the threshold test by combining the protected identity connection, the relevant audience, unprompted recognition, acoustic evidence, and contextual attribution, while remaining analytically distinct from authorisation, justification, fault, causation, and remedy. Responsibility should then follow technical control, understood as the auditable capacity to prevent, trace, constrain, or terminate a defined risk at a particular system node. Chinese rules on personal information, deep synthesis, and AI-content labelling demonstrate how provenance, scoped consent, model documentation, access controls, logs, labels, and complaint mechanisms can translate legal duties into technical architecture. The resulting framework offers a portable but limited technolaw method that restores control where defendants can practically act while preserving commentary, parody, art, accessibility, and innovation.