[Digital Today reporter Jae-won Choi (최재원)] China’s Supreme People’s Court has drawn up standards for judging civil liability stemming from AI deepfakes, voice cloning and AI hallucinations. Using or disclosing an identifiable face or voice replicated by AI without a person’s consent may be deemed an infringement of personality rights.

Cryptopolitan reported on Sept. 7 local time that the Supreme People’s Court issued “Opinions on hearing AI-related dispute cases in accordance with the law” that day. It is judicial guidance for applying the existing Civil Code and the Personal Information Protection Law, among others, to AI disputes.

The key issue is whether the party has consented. Creating a virtual image modeled on a specific person’s face without consent, or training a voice to generate identifiable synthetic speech, can lead to liability for infringing portrait rights and voice rights. Creating false remarks using deepfakes that lower social evaluation also constitutes an infringement of reputation rights.

If harm is severe and expected to be difficult to remedy, a party may also apply to a court for an injunction. A typical case is distributing fabricated sexual content created by compositing a person’s face.

The guidance also sets out standards for liability of AI service operators. If an AI system generates false or inaccurate information that infringes another person’s personality rights, and the operator fails to take necessary steps such as blocking generation after receiving notice from the rights holder, the operator may be liable for damages. If a user created infringing content using malicious prompts, the user is also subject to liability.

It also regulates algorithm-based price discrimination. It allows liability for operators that cause harm by applying different transaction terms to consumers for the same product or service without a reasonable reason.

China’s authorities are also stepping up enforcement against AI content. The Cyberspace Administration of China said on Sept. 2 it had cracked down on AI false information and impersonation, handling more than 5.61 million items of illegal and harmful information and about 49,000 accounts, and taking action against about 2,400 illegal websites and apps.