Ms voice deepfake — Sep 2026
The Shanghai First Intermediate People’s Court ruled against a company for unauthorized AI voice cloning, establishing a new burden-shifting framework for.
- Reported date
- Sep 30, 2026
- Target
- Ms. Wang
The exact incident date was not established. This entry is dated by its source report.
In a landmark ruling, the Shanghai First Intermediate People’s Court held a company liable for using an AI-synthesized version of a voice actor’s voice without authorization. The case establishes a critical evidentiary framework for addressing the information asymmetry inherent in generative AI, shifting the burden of proof regarding training data from the plaintiff to the model operator.
What happened
In January 2025, a voice actor named Ms. Wang discovered that Company A had used an AI-synthesized version of her voice in an internet promotional campaign. Wang, who publishes her work on an app, had never authorized the company to use her voice. She notarized the audio and filed a lawsuit seeking damages. Company A admitted the audio was generated by its own system but claimed it could not verify the training material due to staff turnover, arguing there was no direct evidence her voice was used.
To support her claim, Wang provided a forensic acoustic comparison between the disputed audio and a live sample of her voice. The analysis, covering 28 formant acoustic indicators, showed an overall similarity of 90%. The court ruled that using a person’s voice as training corpus without consent to generate a recognizable mimicry constitutes an infringement of personality rights, regardless of the technical method used.
Crucially, the court addressed the "black box" problem of generative AI. It ruled that a plaintiff meets their burden by showing a plausible channel for data acquisition—such as the plaintiff’s work being publicly available—and demonstrating high similarity between the voices. Once this threshold is met, the burden shifts to the defendant to prove the lawful origin of its training data. Because Company A could not provide records to prove its training material was legitimate, it bore the adverse consequences of that failure. The court ultimately upheld an award of RMB 50,000, noting that while the infringement was clear, the specific use of the voice as a notification prompt and the modest market profile of the plaintiff limited the damages.