August 01, 2025
Introduction to the Measures for Labeling AI-Generated or Composed Content (Mainland China)
On March 7, 2025, the Cyberspace Administration of China, the Ministry of Industry and Information Technology, the Ministry of Public Security, and the National Radio and Television Administration jointly formulated and released the “Measures for Labeling AI-Generated or Composed Content” (hereinafter referred to as the "Measures"), which will come into effect on September 1, 2025. As the implementation of the Measures is imminent, it is necessary to briefly review the main content of the Measures once again in order to reduce the data compliance risks for relevant Internet information service providers:
I. Applicable Entities
According to Article 2 of the Measures, internet information service providers (hereinafter referred to as "service providers") that meet the circumstances stipulated in the Regulations on the Management of Internet Information Service Algorithmic Recommendation, the Regulations on the Management of Deep Synthesis of Internet Information Service, and the Interim Measures for the Management of Generative AI Services are subject to the Measures when conducting activities to identify AI-generated or composed content.
In other words, the so-called "Service Providers" include: service providers offering internet information content dissemination services [1] ; service providers using algorithmic technologies such as generative or composed, personalized recommendation, sorting and selection, search and filtering, and scheduling decision-making to provide information to users [2] ; service providers using deep learning, virtual reality, and other generative or composed algorithms to create text, images, audio, video, virtual scenes, and other internet information [3] ; organizations and individuals providing generative AI services (including through programmable interfaces) using generative AI technology [4] .
II. Objects of Labeling
According to Article 3 of the Measures, AI-generated or composed content refers to text, images, audio, video, virtual scenes, and other information generated or composed using AI technology. That is, whether the content is generated or composed, whether it is text, images, audio, video, or virtual scenes, all are subject to labeling requirements.
III. Methods of Labeling
Article 3 of the Measures also points out that label includes two methods: explicit label and implicit label. Explicit label refers to the label added in the generated or composed content or interactive scene interface, presented in the form of text, sound, graphics, etc., and can be clearly perceived by users; implicit label refers to the label added in the file data of the generated or composed content through technical measures, which is not easily perceived by users. In addition, Articles 4 and 5 of the Measures also specify the circumstances that require explicit and implicit labels:
A. Circumstances Requiring Implicit Labels
In accordance with Article 16 of the Regulations on the Management of Deep Synthesis of Internet Information Service, deep synthesis service providers shall take technical measures to add marks that do not affect users' use to the information content generated or edited using their services, and preserve log information in accordance with laws, administrative regulations, and national provisions. That is, they shall add implicit label in the file metadata of the generated or composed content, which refers to the descriptive information embedded in the file header as per a specific encoding format, used to record the file's source, attributes, purpose, and other information content, in accordance with the requirements of the Measures. The implicit label includes information on the attributes of the generated or composed content, the name or code of the service provider, content identification number, and other production elements.
In addition, the Measures also encourage service providers to add implicit label in the form of digital watermarks to the generated or composed content.
B. Circumstances Requiring Explicit Labels
If the synthetic service provided by the service provider falls under the first paragraph of Article 17 of the Regulations on the Management of Deep Synthesis of Internet Information Service, explicit labels shall be added to the generated or composed content, that is, if the deep synthetic service provider offers deep synthetic services that may lead to public confusion or misidentification, it shall prominently mark the generated or edited content at a reasonable position or area to alert the public to the synthetic nature of the content, including:
1. Services for generating or editing text through intelligent dialogue, intelligent writing, and other means that simulate human text generation or editing;
2. Services for generating or significantly altering personal identity features through synthetic voice, voice imitation, and other voice generation or editing services;
3. Services for generating or significantly altering personal identity features through face generation, face replacement, face manipulation, posture manipulation, and other image or video generation or editing services;
4. Services for generating or editing immersive realistic scenes;
5. Other services with the function of generating or significantly altering the content of information.
The so-called reasonable position or area, according to Article 4 of the Measures, refers to the following positions or areas that meet the following requirements:
1. Adding text prompts or universal symbol prompts as labels at the beginning, end, or appropriate middle position of the text, or adding prominent prompts at the interactive scene interface or around the text;
2. Adding voice prompts or audio rhythm prompts as labels at the beginning, end, or appropriate middle position of the audio, or adding prominent prompts at the interactive scene interface;
3. Adding prominent prompts as labels at appropriate positions on the image;
4. Adding prominent prompts as labels at appropriate positions around the video playback and at the beginning of the video, and it is also possible to add prominent prompts at appropriate positions in the middle and end of the video;
5. When presenting virtual scenes, adding prominent prompts as labels at appropriate positions on the initial screen, and it is also possible to add prominent prompts at appropriate positions during the continuous service of the virtual scene;
6. In other synthetic service scenarios, adding prominent prompts as labels according to the application characteristics of the service.
In addition, when service providers offer functions for downloading, copying, exporting, or other similar actions of generated or composed content, they shall ensure that the files contain explicit labels that meet the requirements.
IV. Main Compliance Obligations
According to the Measures and the Compliance Guidelines for Generative AI Service Enterprises, in addition to service providers, providers of internet information content dissemination services, internet application distribution platforms, and ordinary users also have certain labeling compliance obligations, as follows:
A. Service Providers (excluding Providers of internet Information Content Dissemination Services)
- Add explicit or implicit labels in accordance with the aforementioned provisions.
- Clearly specify the methods and styles of labels used in generated or composed content in the user service agreement, and remind users to carefully read and understand the relevant label management requirements.
- When users request to provide generated or composed content without explicit labels, legally retain logs of the recipient's information for no less than six months.
- Comply with relevant laws, administrative regulations, departmental rules, and mandatory national standards.
- When fulfilling procedures such as algorithm filing and security assessment, provide materials related to labels used in generated or composed content, and enhance the sharing of labeling information to support and assist in preventing and combating related illegal and criminal activities.
B. Providers of internet Information Content Dissemination Services
- Check whether the content contains implicit labels.
- Directly mark content that contains implicit labels or contains a declaration.
- Mark content that is suspected to be generated or composed.
- Provide labeling functions and guide users to mark content.
C. Internet Application Distribution Platforms
- Require an explanation of whether the application has generating or composing functions and conduct corresponding checks during the application's listing or launch review.
D. Users
- When using internet information content dissemination services to publish generated or composed content, they shall proactively declare and use the labeling functions provided by the service provider to mark the content.
In addition, no one shall maliciously delete, alter, forge, or conceal the labels of generated or composed content stipulated in the Measures, nor shall they provide tools or services for others to carry out such malicious acts, nor shall they use improper labeling means to infringe upon the legitimate rights and interests of others.
Although the Measures do not specify clear legal liabilities, any violation of the provisions of the Measures shall be handled by the relevant competent authorities such as cyberspace administration, telecommunications, public security, and radio and television authorities in accordance with relevant laws, administrative regulations, and departmental rules. It is recommended that all types of service providers promptly conduct self-inspections to identify and fill gaps to avoid compliance risks.
[1] Article 6 of the Measures states that providers of internet information content dissemination services shall take measures to regulate the dissemination activities of generated or composed content, indicating that providers of internet information content dissemination services are also subject to the management of the Measures in addition to the aforementioned three regulations.
[2] See Article 2 of the Regulations on the Management of Internet Information Service Algorithmic Recommendation, which stipulates that the application of algorithmic recommendation technology to provide Internet information services (hereinafter referred to as algorithmic recommendation services) within the territory of the People's Republic of China is subject to these regulations. If there are other provisions in laws and administrative regulations, those provisions shall apply.
The term "application of algorithmic recommendation technology" referred to in the preceding paragraph means the use of algorithmic technologies such as generative or composed, personalized recommendation, sorting and selection, search and filtering, and scheduling decision-making to provide information to users.
[3] See Article 23 of the Regulations on the Management of Deep Synthesis of Internet Information Service, which defines deep synthesis technology as the use of deep learning, virtual reality, and other generative or composed algorithms to create text, images, audio, video, virtual scenes, and other internet information.
[4] See Paragraph (2), Article 22 of the Interim Measures for the Management of Generative AI Services, which defines a generative AI service provider as an organization or individual providing generative AI services (including through programmable interfaces) using generative AI technology.
I. Applicable Entities
According to Article 2 of the Measures, internet information service providers (hereinafter referred to as "service providers") that meet the circumstances stipulated in the Regulations on the Management of Internet Information Service Algorithmic Recommendation, the Regulations on the Management of Deep Synthesis of Internet Information Service, and the Interim Measures for the Management of Generative AI Services are subject to the Measures when conducting activities to identify AI-generated or composed content.
In other words, the so-called "Service Providers" include: service providers offering internet information content dissemination services [1] ; service providers using algorithmic technologies such as generative or composed, personalized recommendation, sorting and selection, search and filtering, and scheduling decision-making to provide information to users [2] ; service providers using deep learning, virtual reality, and other generative or composed algorithms to create text, images, audio, video, virtual scenes, and other internet information [3] ; organizations and individuals providing generative AI services (including through programmable interfaces) using generative AI technology [4] .
II. Objects of Labeling
According to Article 3 of the Measures, AI-generated or composed content refers to text, images, audio, video, virtual scenes, and other information generated or composed using AI technology. That is, whether the content is generated or composed, whether it is text, images, audio, video, or virtual scenes, all are subject to labeling requirements.
III. Methods of Labeling
Article 3 of the Measures also points out that label includes two methods: explicit label and implicit label. Explicit label refers to the label added in the generated or composed content or interactive scene interface, presented in the form of text, sound, graphics, etc., and can be clearly perceived by users; implicit label refers to the label added in the file data of the generated or composed content through technical measures, which is not easily perceived by users. In addition, Articles 4 and 5 of the Measures also specify the circumstances that require explicit and implicit labels:
A. Circumstances Requiring Implicit Labels
In accordance with Article 16 of the Regulations on the Management of Deep Synthesis of Internet Information Service, deep synthesis service providers shall take technical measures to add marks that do not affect users' use to the information content generated or edited using their services, and preserve log information in accordance with laws, administrative regulations, and national provisions. That is, they shall add implicit label in the file metadata of the generated or composed content, which refers to the descriptive information embedded in the file header as per a specific encoding format, used to record the file's source, attributes, purpose, and other information content, in accordance with the requirements of the Measures. The implicit label includes information on the attributes of the generated or composed content, the name or code of the service provider, content identification number, and other production elements.
In addition, the Measures also encourage service providers to add implicit label in the form of digital watermarks to the generated or composed content.
B. Circumstances Requiring Explicit Labels
If the synthetic service provided by the service provider falls under the first paragraph of Article 17 of the Regulations on the Management of Deep Synthesis of Internet Information Service, explicit labels shall be added to the generated or composed content, that is, if the deep synthetic service provider offers deep synthetic services that may lead to public confusion or misidentification, it shall prominently mark the generated or edited content at a reasonable position or area to alert the public to the synthetic nature of the content, including:
1. Services for generating or editing text through intelligent dialogue, intelligent writing, and other means that simulate human text generation or editing;
2. Services for generating or significantly altering personal identity features through synthetic voice, voice imitation, and other voice generation or editing services;
3. Services for generating or significantly altering personal identity features through face generation, face replacement, face manipulation, posture manipulation, and other image or video generation or editing services;
4. Services for generating or editing immersive realistic scenes;
5. Other services with the function of generating or significantly altering the content of information.
The so-called reasonable position or area, according to Article 4 of the Measures, refers to the following positions or areas that meet the following requirements:
1. Adding text prompts or universal symbol prompts as labels at the beginning, end, or appropriate middle position of the text, or adding prominent prompts at the interactive scene interface or around the text;
2. Adding voice prompts or audio rhythm prompts as labels at the beginning, end, or appropriate middle position of the audio, or adding prominent prompts at the interactive scene interface;
3. Adding prominent prompts as labels at appropriate positions on the image;
4. Adding prominent prompts as labels at appropriate positions around the video playback and at the beginning of the video, and it is also possible to add prominent prompts at appropriate positions in the middle and end of the video;
5. When presenting virtual scenes, adding prominent prompts as labels at appropriate positions on the initial screen, and it is also possible to add prominent prompts at appropriate positions during the continuous service of the virtual scene;
6. In other synthetic service scenarios, adding prominent prompts as labels according to the application characteristics of the service.
In addition, when service providers offer functions for downloading, copying, exporting, or other similar actions of generated or composed content, they shall ensure that the files contain explicit labels that meet the requirements.
IV. Main Compliance Obligations
According to the Measures and the Compliance Guidelines for Generative AI Service Enterprises, in addition to service providers, providers of internet information content dissemination services, internet application distribution platforms, and ordinary users also have certain labeling compliance obligations, as follows:
A. Service Providers (excluding Providers of internet Information Content Dissemination Services)
- Add explicit or implicit labels in accordance with the aforementioned provisions.
- Clearly specify the methods and styles of labels used in generated or composed content in the user service agreement, and remind users to carefully read and understand the relevant label management requirements.
- When users request to provide generated or composed content without explicit labels, legally retain logs of the recipient's information for no less than six months.
- Comply with relevant laws, administrative regulations, departmental rules, and mandatory national standards.
- When fulfilling procedures such as algorithm filing and security assessment, provide materials related to labels used in generated or composed content, and enhance the sharing of labeling information to support and assist in preventing and combating related illegal and criminal activities.
B. Providers of internet Information Content Dissemination Services
- Check whether the content contains implicit labels.
- Directly mark content that contains implicit labels or contains a declaration.
- Mark content that is suspected to be generated or composed.
- Provide labeling functions and guide users to mark content.
C. Internet Application Distribution Platforms
- Require an explanation of whether the application has generating or composing functions and conduct corresponding checks during the application's listing or launch review.
D. Users
- When using internet information content dissemination services to publish generated or composed content, they shall proactively declare and use the labeling functions provided by the service provider to mark the content.
In addition, no one shall maliciously delete, alter, forge, or conceal the labels of generated or composed content stipulated in the Measures, nor shall they provide tools or services for others to carry out such malicious acts, nor shall they use improper labeling means to infringe upon the legitimate rights and interests of others.
Although the Measures do not specify clear legal liabilities, any violation of the provisions of the Measures shall be handled by the relevant competent authorities such as cyberspace administration, telecommunications, public security, and radio and television authorities in accordance with relevant laws, administrative regulations, and departmental rules. It is recommended that all types of service providers promptly conduct self-inspections to identify and fill gaps to avoid compliance risks.
[1] Article 6 of the Measures states that providers of internet information content dissemination services shall take measures to regulate the dissemination activities of generated or composed content, indicating that providers of internet information content dissemination services are also subject to the management of the Measures in addition to the aforementioned three regulations.
[2] See Article 2 of the Regulations on the Management of Internet Information Service Algorithmic Recommendation, which stipulates that the application of algorithmic recommendation technology to provide Internet information services (hereinafter referred to as algorithmic recommendation services) within the territory of the People's Republic of China is subject to these regulations. If there are other provisions in laws and administrative regulations, those provisions shall apply.
The term "application of algorithmic recommendation technology" referred to in the preceding paragraph means the use of algorithmic technologies such as generative or composed, personalized recommendation, sorting and selection, search and filtering, and scheduling decision-making to provide information to users.
[3] See Article 23 of the Regulations on the Management of Deep Synthesis of Internet Information Service, which defines deep synthesis technology as the use of deep learning, virtual reality, and other generative or composed algorithms to create text, images, audio, video, virtual scenes, and other internet information.
[4] See Paragraph (2), Article 22 of the Interim Measures for the Management of Generative AI Services, which defines a generative AI service provider as an organization or individual providing generative AI services (including through programmable interfaces) using generative AI technology.
