Shanghai First Intermediate People’s Court Holds AI Voice-Cloning Platform Liable, Shifts Burden of Proof on Training Data

This ruling establishes a critical evidentiary framework for proving AI voice infringement when training data is opaque. By shifting the burden of proof to the defendant, it allows plaintiffs to rely on forensic similarity and public accessibility rather than requiring direct evidence of data sourcing. This approach directly addresses the structural information asymmetry inherent in generative AI systems, ensuring that individuals can protect their personality rights despite the "black box" nature of model training. The decision underscores the legal principle that recognizing a synthetic voice as a specific individual constitutes infringement, regardless of the underlying algorithmic method. This clarifies that personality interests in voice are protected against appropriation, even when the connection between original recordings and AI outputs is not explicitly documented. It emphasizes that the focus must remain on the recognizability of the output to the public, rather than the technical specifics of how the AI was trained. For open data and AI development, this case highlights the growing legal risks associated with using publicly available data for model training without explicit consent. It signals a move toward stricter accountability, requiring AI providers to maintain transparent records of their training corpora to defend against infringement claims. Consequently, organizations must carefully manage data sourcing and consent mechanisms to avoid liability, impacting how open datasets involving personal voices are collected, stored, and utilized in generative models.

Source: natlawreview.com
Published on 2026-10-08

Shanghai First Intermediate People’s Court Holds AI Voice-Cloning Platform Liable, Shifts Burden of Proof on Training Data

Related news