On September 7, China's Supreme People's Court released the Opinions on the Lawful Trial of Cases Involving AI-Related Disputes, a document with five parts and 24 articles in total. This marks the first time the country's top judicial body has consolidated adjudication standards for AI-related cases into a single dedicated document.
Start with the legal weight of the document. This is an "opinion," not a judicial interpretation with binding legal effect of its own. It guides courts at all levels, but judges still cite the Personality Rights and Tort Liability books of the Civil Code and the Consumer Rights Protection Law when writing rulings — the opinion's role is to align how different courts read the same new types of conduct. Over the past two years, internet courts in Hangzhou and Beijing have ruled on AI face-swap and voice-cloning cases, but courts have not been fully consistent on whether liability extends to the training stage or how damages should be calculated.
Liability for voice cloning shifts earlier
The section on personality rights is fairly specific: using AI without consent to process a natural person's name, likeness or other attributes to generate and use or publish a recognizable virtual digital persona constitutes an infringement of personality rights such as the right to name and the right to likeness.
Using a natural person's voice as training data without consent, and generating a recognizable synthetic voice by imitating their timbre, intonation and speaking style, constitutes an infringement of voice rights. The point of liability shifts from "does the output sound like the person" to "was your voice used during training" — a change that directly affects voice actors, livestreamers and others whose livelihood depends on their voice. The opinion also brings AI-enabled doxxing and "human flesh search" within the scope of privacy-rights protection.
Algorithmic price discrimination and celebrity-impersonation endorsements
For the same goods or services, an operator that uses algorithms to apply unreasonable differential treatment causing harm must bear corresponding tort liability. This provision pulls "big data killing familiar customers" — charging loyal users more — out of platform self-regulation and into the scope of civil tort law.
Cases of AI impersonating celebrities to endorse products are treated more severely. Where AI is used to impersonate someone else's endorsement to induce consumption and this constitutes fraud, consumers' claims for punitive damages can be supported under the law. Punitive damages in consumer cases are calculated as a refund plus triple compensation, meaning the payout triggered by a few dozen seconds of counterfeit video could far exceed the cost of producing it.
Self-driving cars: liability can run both ways
The section on autonomous driving gives victims a parallel path: where harm results from a combination of a vehicle defect and driver fault, a victim who simultaneously claims against both the driver and the manufacturer or seller will be supported under the law. Another article addresses marketing: false claims about the level of automation or performance entitle consumers to civil compensation.
Many Chinese automakers market Level 2 driver-assistance systems as "smart driving" or even "autonomous driving." This rule gives car buyers a clear basis for a claim.
Five openings on intellectual property
Part three, Articles 12 through 16, covers five areas: determining tort liability, legal liability for open-source software, patent authorization and confirmation, performance of technology contracts, and data-use norms. Open-source software has long lacked stable judicial precedent in China — questions like what license model weights and training scripts should be distributed under, and whether violating a license counts as breach of contract or infringement, have mainly been handled under general contract-law rules until now.
Part five, Articles 21 through 24, covers the courts' own working mechanisms, including dispute prevention and resolution, research, guiding cases and adjudication oversight, as well as inter-agency coordination and international judicial exchange and cooperation.
For ordinary people, the first two sections are the most immediately useful. If your face or voice has been used to generate content without consent, the evidence-gathering task now extends beyond "finding the video" to "proving the other party used your material for training" — a much harder technical bar. The opinion itself does not specify how the burden of proof should be allocated; how this plays out in individual cases will depend on guiding cases issued later.
Sources: Supreme People's Court's Opinions on the Lawful Trial of Cases Involving AI-Related Disputes, China News Service, CocoLoop, Jiemian News, IT Home; article counts and section breakdown verified against the text published by the Supreme People's Court.