Explicit and implicit labels
The Measures for Labelling AI-Generated Synthetic Content, issued in March 2025 by the Cyberspace Administration of China with three other agencies, took effect on 1 September 2025 alongside the mandatory national standard GB 45438-2025. They distinguish explicit labels, which people can see or hear, such as text on an image, a notice at the start of a video or an audio prompt, from implicit labels embedded in file metadata. Platforms must look for both, add labels where they detect synthetic content and ask users to declare AI-generated material when they post it. Maliciously removing, altering or forging labels is prohibited. The earlier Deep Synthesis Provisions, in force since January 2023, already required conspicuous labels on face swaps, synthetic voices and other content that could mislead.
Faces, voices and personal information
Under the Personal Information Protection Law, in force since November 2021, facial and voice data used to identify people is sensitive personal information, which needs a specific purpose, necessity and separate consent. The Deep Synthesis Provisions add that anyone offering face or voice editing must prompt users to obtain the consent of the person depicted. The Civil Code prohibits forging someone's portrait with information technology and extends the same protection to voices, and the Beijing Internet Court applied that to an AI voice built from a voice actor's recordings in 2024. Moving source recordings out of China triggers the cross-border transfer rules, so decide early where likeness data is captured, stored and processed.
Generative services, content lines and maps
The Interim Measures for generative AI services, in force since August 2023, apply to services offered to the public in China: providers must meet content requirements and, where a service can shape public opinion, complete a security assessment and algorithm filing. In practice, brands producing at scale for China often work through filed domestic services or local partners, and every asset passes platform review before it runs. Content lines are stricter than elsewhere on politics, history, religion and national symbols. Maps are a specific trap: any map in published material must show China's borders exactly as officially approved, and a generated globe, a stylised world map or a background atlas can fail on a detail no one in the room noticed.
Superlatives, platforms and script
The Advertising Law forbids terms such as 'national-level', 'highest' and 'best', with fines from RMB 200,000 to 1 million, and regulators enforce it. Generated copy reaches for superlatives by default, so they belong on a banned-word list rather than in a reviewer's notes. The channels are different too. Content runs on WeChat, Douyin, Xiaohongshu, Weibo, Bilibili and Kuaishou, and commerce on Tmall, JD.com and Pinduoduo, each with its own formats and review. Mainland audiences read simplified characters; Hong Kong, Macau and Taiwan use traditional characters and different vocabulary, and Hong Kong voice-overs are usually Cantonese. The studio's brand hub enforces banned words at generation, so a superlative never reaches the approval queue in the first place.
Updated 25 September 2026 · General information, not legal advice. Rules change, so check the current text with your legal team before relying on it.