AINEWS
The opinion treats likeness, voice, and personal traits as civil interests and covers deepfakes, doxing, and other AI harms. Lower Chinese courts must refer to it, with advisory force in a civil law system.
China's Supreme People's Court didn't ban deepfakes. In a 24-article opinion released Monday, the court reclassified likeness, voice, and personal traits as protectable civil interests. It told lower courts to refer to the list when adjudicating AI disputes ranging from voice cloning to doxing (the public release of someone's private information), per the SCIO/Xinhua press release.
Si Yanli, deputy head of the SPC's research office, framed the document around the protection of individual identities, language, and personal traits. The opinion walks through specific examples: deepfakes used to "bring the dead back to life," and cases in which a person's face or voice is treated as material that can be "harvested" or "stolen" by a model, language echoed in the Futurism re-report and the SPC's own English summary. Procedurally, the opinion says even a single AI-generated frame of someone's likeness can be enough to ground a civil case.
AI-generated video and audio lead the 24 articles, followed by sexualized deepfake content, financial misconduct enabled by synthetic media, digital privacy, doxing, and surveillance pricing (the commercial trade in monitoring data). These are civil remedies, not new crimes. The "crackdown" framing that has followed the release collapses that distinction because the underlying mechanism is structural. The PRC lacks the common-law doctrine of stare decisis, the rule that earlier court decisions bind later ones, so an SPC opinion does not bind lower courts the way a US Supreme Court ruling would. The original Chinese opinion, with a full English translation by China Law Translate, is structured as guidance, not a verdict.
Lower Chinese courts are required to refer to SPC opinions when adjudicating disputes, which is where the document gets its practical force. A judge handling a voice-cloning case in 2027 will read the same 24 articles her colleague in another province read in 2026. The opinion functions as a national reference set, a checklist of the harm categories Chinese courts are expected to recognize and the remedies they are expected to apply. For non-Chinese regulators drafting deepfake rules, that checklist is now a public comparison document.
The opinion does not address whether AI-generated works are themselves copyrightable, an omission flagged in National Law Review's legal analysis. The court's framing treats the harm as something done to a person, not something done by a model, which leaves the underlying output question unresolved. For platforms and model providers operating in or exporting AI services into the PRC, that split narrows the surface for personal-likeness liability while leaving the property-rights layer for another ruling. A US model host whose outputs are scraped and republished inside China now has a clearer pathway to a civil suit from the depicted individual, but a fuzzier route to a claim against the downstream re-publisher of the model output itself.
The doxing and sexualized-content clauses are being read as a meaningful expansion of personal protection, the first time a top Chinese court has enumerated these categories at the national level. Critics note the same doctrinal flexibility can extend state surveillance, since "surveillance pricing" appears in the document as a covered category. The 24 articles do not draw that line themselves.
A Beijing intermediate court earlier this year ruled against a company that used AI to liquidate an employee's position, and the SPC's national guidance gives that line of reasoning a ready-made citation. Within months, lower courts will be the ones deciding how the 24 articles actually apply. The first dozen rulings under the new framework will show whether the document reads as personal protection, as state expansion, or as the boundary case the SPC left for the rest of the system to redraw.