AI-generated deepfakes of celebrities violate rights in China
AI systems generated deepfake videos of Chinese celebrities including Zhou Xingchi, Feng Gong, Cai Ming and Liu Dehua without authorization in February 2026. Legal experts stated the synthetic content violated portrait rights and personality rights under China's civil code, and could facilitate fraud and misinformation despite platform inaction.
What happened
AI-generated videos of actor Zhou Xingchi recreating scenes from his classic films were widely shared on a short video platform in early February 2026. Zhou's agent Chen Zhenyu posted questioning whether the content constituted copyright infringement, noting that creators appeared to be profiting while the platform failed to take action.
Around the same time, unauthorized AI-generated Lunar New Year greeting videos featuring comedians Feng Gong and Cai Ming, as well as actor Liu Dehua, circulated on social media platforms, showing them in domestic scenes wishing viewers good fortune.
Reported impact
According to Zhang Linghan, director of the Artificial Intelligence Law Research Institute at China University of Political Science and Law, faces and voices are sensitive personal information with biometric functions. Unauthorized deepfake synthesis infringes on individuals' personality rights and portrait rights.
Experts noted that highly realistic content can blur the boundaries between real and fake, potentially facilitating fraud and rumors that damage the online ecosystem. Legal commentary stated that even non-commercial use of such deepfakes without authorization violates law under China's civil code and deepfake management regulations.
Index analysis
The incidents highlight the technical and legal gaps in AI content generation governance. While Google's Gemini and Nano Banana systems can be programmed to reject direct keyword requests for protected characters, testing revealed they still generate content when users upload reference images combined with text prompts, exposing a monitoring gap.
Industry observers noted that AI model training relies on vast datasets whose copyright ownership is often unclear. The incidents underscore the contradiction between AI's data requirements and intellectual property protection in AIGC development.
Governance controls
Media commentary called for platforms to strengthen content review and labeling requirements, while urging users to develop copyright awareness and refuse to create or share infringing content. Experts stated that the boundary for technological innovation must be respected and that rules cannot be subordinated to entertainment value.
Industry participants have begun exploring licensing partnerships combined with technical controls as a model to reduce copyright litigation risk. Google stated its AI models use publicly available training data, while Disney simultaneously reached a $1 billion licensing agreement with OpenAI for use of Marvel and Lucasfilm intellectual property in its Sora video generation tool.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.