--- title: "China now forces a label on every AI-made image, video and text" date: 2026-09-29 category: Policy & Governance site: NeuroAI canonical: https://neuroai.site/a/na-policy-aigc-labeling-cn language: en --- # China now forces a label on every AI-made image, video and text > Since 1 September 2025, China has required service providers to mark AI-generated and AI-synthesized content — explicitly on the surface and implicitly in file metadata. The Measures for the Labeling of AI-Generated and AI-Synthesized Content (人工智能生成合成内容标识办法), issued by four central agencies, are now a year old and shape how every Chinese platform handles synthetic media. A photo that looks real but was never shot. A voice that sounds like someone who never spoke. China decided, a year ago, that both should carry a tag you can see — and one you usually cannot. ## The rule that actually took effect The Measures for the Labeling of AI-Generated and AI-Synthesized Content (人工智能生成合成内容标识办法) were issued on 7 March 2025 by four agencies — the Cyberspace Administration of China (CAC, 国家网信办), the Ministry of Industry and Information Technology, the Ministry of Public Security, and the National Radio and Television Administration — and took effect on 1 September 2025. The measure sits on top of earlier rules: the Interim Measures for the Management of Generative AI Services (生成式人工智能服务管理暂行办法), in force since 15 August 2023, and the Deep Synthesis Provisions (互联网信息服务深度合成管理规定). It is the first to make labeling a standing obligation rather than guidance. ## Two kinds of label The rule splits marking into two types: - **Explicit labels (显式标识)** — text, sound or graphics a user can perceive, placed on the content or in the interaction interface. Text gets a notice at the start, end or middle; images get a visible mark; video gets a prompt at the start and around playback; virtual scenes get a mark at the opening. - **Implicit labels (隐式标识)** — technical metadata baked into the file recording the content's nature, the service provider's name or code, and a content ID. Digital watermarks are encouraged. Crucially, the duty does not stop at the generator. Platforms that transmit content must check metadata, and if they find synthetic content — or detect an explicit mark, or a user declares it — they must add a visible prompt at the post. App stores must verify labeling materials before listing an AI service. ## Why it landed when it did By the time the rule took effect, China's generative-AI base was already large. State broadcaster CCTV reported that as of 1 September 2025, more than 490 large models had been filed with the CAC, over 240 had been registered with provincial cyberspace offices, and generative-AI users had reached 230 million. A companion mandatory national standard — GB 45438-2025, "Cybersecurity Technology — Labeling Method for AI-Generated Synthetic Content" — also took effect on 1 September 2025, giving engineers a concrete format for the metadata and watermark fields. ## How platforms are responding In the year since effect, major Chinese apps and model services have added visible "AI-generated" tags on outputs and embedded metadata on export. The CAC has run local briefings in nine provinces covering more than 700 organizations, according to its own account, to walk providers through the standard. The practical effect: a synthetic clip now arrives with a mark, and a platform that strips it faces a prompt to re-label on posting. ## What it does not do The measure is a labeling and traceability rule, not a ban. It tells providers to mark and keep records; it does not prohibit synthetic content. It also stops short of judging whether a given output is true or false — it only requires that its machine origin be declared. Users who maliciously delete, alter, forge or hide a required label can be pursued under the relevant laws. ## What readers can do now - **If you publish on Chinese platforms**, assume every AI-generated post needs a visible mark and clean metadata; strip-and-repost workflows will now trigger platform prompts. - **If you build AI tools**, implement explicit labels at export and implicit metadata per GB 45438-2025 before launch — app-store review now checks for it. - **If you follow AI governance**, treat 1 September 2025 as the date China moved from "register the model" to "label the output" as its core compliance step. ## Honest limitations Core facts: the Measures for the Labeling of AI-Generated and AI-Synthesized Content (人工智能生成合成内容标识办法) issued 7 March 2025 by CAC, MIIT, Ministry of Public Security and NRTA; effective 1 September 2025; two label types (explicit 显式标识 / implicit 隐式标识); transmission platforms and app stores carry duties; companion mandatory standard GB 45438-2025 effective 1 September 2025; as of that date, 490+ large models filed with CAC, 240+ registered provincially, 230 million generative-AI users (figures from CCTV, 1 September 2025). Sourced from the official CAC notice (cac.gov.cn, 14 March 2025 publication), the State Council portal (gov.cn), and CCTV. The 490/240/230-million numbers are CCTV's reported tallies as of the effective date, not an independent audit. The umbrella rules (生成式人工智能服务管理暂行办法, effective 15 August 2023) and Deep Synthesis Provisions are correctly cited as prior framework. This article contains no RMB amounts, so no currency conversion applies. Coverage is of labeling mechanics and effective dates only, with no political commentary; analysis is current to 29 September 2026. --- Published by NeuroAI (https://neuroai.site/) — https://neuroai.site/a/na-policy-aigc-labeling-cn Free to quote with attribution and a link to the original.