The Supreme People's Court of China issued the Opinions on Lawfully Adjudicating Disputes Involving Artificial Intelligence on 7 September 2026, numbered Fafa [2026] No. 10. The Opinions are a judicial policy document addressed to courts at every level and took effect on issue. They contain 24 articles arranged in five parts covering general requirements, infringement liability, intellectual property, procedural and criminal rules, and court working mechanisms.
Article 4 treats the processing of a natural person's name or likeness to generate and publish a recognisable virtual digital image without that person's consent as an infringement of personality rights, and applies the same treatment to using a person's voice as training material to imitate their timbre and intonation and produce a synthesised voice that identifies them. Article 6 provides that processing personal information the individual has published themselves, or that has otherwise been lawfully made public, within a reasonable scope for model training is generally not an infringement of personal information rights where the individual has not expressly objected. Article 7 places liability on a generative AI service provider that, after notice from the rights holder, fails to take necessary measures promptly, including ceasing generation of the infringing content.
Generative AI service providers operating in China face a notice and action standard that the courts will apply directly. Model operators, including domestic large language model providers and the platforms distributing their output, need a channel to receive a rights holder notice and the technical ability to stop generation of the offending content rather than only delete a stored copy. Voice actors, performers and other individuals whose likeness or voice has been synthesised have a stated route to relief under Article 4. Developers training on scraped public data gain a qualified position under Article 6 that turns on reasonable scope and the absence of an express objection.
Article 6 does not set out how an individual records an objection, so the point at which lawfully public data stops being available for training is left to the trial courts. The Opinions also leave open whether AI generated output attracts copyright protection, which continues to be decided case by case under existing copyright rules. Because the document is issued as a Fafa policy instrument rather than a judicial interpretation, lower courts are directed by it while litigants cannot plead it as the legal basis of a claim.
Licentium advises AI developers, platform operators and rights holders on regulatory exposure across jurisdictions. Work we undertake includes assessing training data provenance against local personal information rules, drafting notice and takedown procedures that meet court expectations, and preparing liability positions for generative AI services distributed into China.