Deep synthesis (“DS”) technology is a widely used application in the field of artificial intelligence (“AI”) and a variety of scenarios, particularly in audio and video production, media communication and information services. For instance, using AI technology to “resurrect” dead people on digital gadgets is becoming a business as shown during the Qingming Festival in 2024. But whether the technology and related business infringe upon the rights and interests of others or public interests is open to question. To address legal and ethical risks and harm posed by DS techniques, China has promulgated a series of laws, regulations and standards, aiming to implement the adaptable and agile governance approach, mitigate AI-related risks and ultimately ensure responsible AI development.
- Key Regulatory Authorities and Fundamental Principles
On November 25, 2022, the Cyberspace Administration of China (“CAC”), the Ministry of Industry and Information Technology (“MIIT”) and the Ministry of Public Security (“MPS”) have jointly issued the Administrative Provisions on Deep Synthesis of Internet Information Services (the “DS Administrative Provisions”), which takes effect on January 10, 2023.
According to the DS Administrative Provisions, the regulatory authorities include: (i) the national cyberspace authority, which is responsible for the overall organization and coordination of the governance and the related supervision and regulation of DS services nationwide, (ii) the telecommunications authority and public security authority under the State Council, which are responsible for the supervision and regulation of DS services. The local counterparts of each regulatory authority are responsible for the overall organization and coordination of the governance or supervision and regulation of DS services within their respective administrative regions.
The DS Administrative Provisions is promulgated based upon the following laws and regulations: (i) the Cybersecurity Law of the People’s Republic of China (the “PRC”), (ii) the Data Security Law of the PRC, (iii) the Personal Information Protection Law of the PRC, (iv) the Administrative Measures on Internet Information Services and other laws and administrative regulations, It can not only be regarded as one of the key components in guaranteeing the healthy and orderly development of the digital economy, but also is in response to the rapid development of AI technology in the new digital era.
For purposes of curbing “deepfake” risks and the risk of violating people’s rights, or even committing crimes, the DS Administrative Provisions require DS service providers (i) to operate their business in accordance with relevant laws and regulations, (ii) to respect social morality and ethics, (iii) to maintain correct political direction, right direction of public opinion as well as correct value orientation, so as to promote the growth of DS services for good.
Chapter II of the DS Administrative Provisions sets out the general rules DS service provides shall comply with, details of which are summarized in the following table:
| No. | What a DS service provider Should Do | What a DS service provider Should Not Do |
| establish and improve management mechanisms and systems in terms of user registration, algorithm review, scientific and technological ethics review, information release review, data security, personal information protection, anti-telecom and online fraud, emergency response; formulate and disclose management rules, platform conventions, user/service agreements | use DS services to produce, reproduce, release or distribute information or engage in activities prohibited by laws or administrative regulations | |
| verify the identity information of DS service users based on mobile phone number, ID card number, unified social credit code or national online identity authentication services | use DS services to produce, reproduce, release or distribute fake news information | |
| strengthen management of DS-produced and employee technical or manual methods to review the data input and synthesis results generated by DS services users | ||
| establish the rumor refutation mechanism as well as convenient access for users to file complaints and the public to submit complaints; deal with complaints and provide feedback in a timely manner |
- Basic Requirements of Legal Safeguard
- Data and information process
Massive data and rich application scenarios serve as the strong advantages of AI development in China. The application data generated by massive users continuously optimizes model performance, forming a positive cycle of “data-algorithm-scenario”.
China has established a sound system of laws to protect people’s rights and interests in cyberspace and built a line of defense in law for protecting personal information rights and interests, which is reflected in (i) the Civil Code, (ii) the Criminal Law (adding provisions on the crime of infringing upon citizens’ personal information), (iii) the Cybersecurity Law, (iv) the Data Security Law, and (v) the Personal Information Protection Law. The DS Administrative Provisions consist of a specific chapter to regulate data-related activities, requiring DS service providers to take necessary measures to ensure the safety of training data and strength the technical management, reviewing, assessing and verifying algorithm mechanism on a regular basis.
In particular, the heart of DS technology lies the application of generative AI to process sensitive personal data, including biometric features, which raises several compliance concerns. Key issues include the sourcing of training data, the risk of generating misleading or false information, and ensuring the security of the data. If AI-generated content misrepresents a target object, like the deceased in the business of “resurrecting” dead people on digital gadgets or is used to deceive others, it could lead to widespread misinformation and potential harm.
Overall, the entire process of collecting, processing, and using data must adhere to the principles of legality, necessity, fairness, and transparency as outlined in the Personal Information Protection Law. The DS Administrative Provisions stipulates that (i) providers of DS services and functions such as editing biometric information should prompt users to inform and obtain consent from individuals whose data is being tampered with, and (ii) the application of DS technology must not infringe on citizens’ rights. In addition, providers of the DS services, such as face-swapping, must provide “prominent identification functions”, which should be able to identify the virtual synthesized “persons”, if the service is “likely to cause public confusion or misidentification”.
- AI-generated content labeling
On September 14, 2024, CAC has released a draft regulation, named the Measures for Labeling AI-generated Synthetic Content (Draft for Comment) (the “Draft Regulations on Labeling AIGC”), aiming to standardize the labeling of AI-generated synthetic content to protect national security and public interests.
AI-generated synthetic content (“AIGC”), as defined by the Draft Regulations on Labeling AIGC, is any text, image, audio or video created using AI technologies. According to the Draft Regulations on Labeling AIGC, (i) internet information service providers shall adhere to mandatory national standards when labeling AIGC. Providers offering functions like downloading, copying or exporting AI-generated materials that may cause confusion or misidentification of the public must ensure that explicit labels are embedded in the files; and (ii) platforms that distribute content are also required to regulate the spread of AI-generated materials by offering identification functions and reminding users to disclose whether their posts contain AIGC.
As a practical matter, requiring service providers to label products is a good way of regulating the DS technique. AIGC-related products should adhere to technical management norms. Relevant industry standards have been published and may be officially introduced to require technology service providers to unify the code when labeling AIGC or even add watermarks to generative content.
- Mandatory algorithm filing
It is well known that personal information security, fake information and algorithm discrimination are problems that have emerged in the development of AI technology. In response to problems raised by algorithm, China has issued laws and regulations, like the Administrative Provisions on Recommendation Algorithms of Internet Information Service (the “Algorithms Administrative Provisions”), to improve and optimize the legality of AI applications.
According to the Algorithms Administrative Provisions, algorithm-based service providers shall comply with relevant laws and regulations, respect social morality and ethics, observe commercial and professional ethics, and follow the principles of fairness and equity, openness and transparency, scientific reasonableness, as well as honesty and credibility.
In addition to (i) information management standards such as adopting measures to ensure information security and (ii) specific rules for protecting users’ rights and interests, the recommendation algorithm-based service providers with public opinion attributes or social mobilization capabilities shall complete the filing procedures within ten business days since it starts offering such services.
Pursuant to the DS Administrative Provisions, the DS service providers with public opinion attributes or social mobilization capabilities shall complete the filing procedures in accordance with the Algorithms Administrative Provisions. CAC has also published a guide to proceeding with such online filing. To date, CAC has released nine batches of algorithms filing Information of DS service providers and over 2,800 algorithms have been filed at CAC.
- Rules to Address Ethical Challenge
AI not only presents enormous opportunities for humanity but also creates unpredictable risks and challenges. Both legal rules and ethical standards are needed to ensure that AI does not pose a threat to humanity. As AI continues to evolve, the ethical implications are becoming an increasingly significant aspect of its regulation. In this process, ethics are increasingly becoming the foundation for legislation, and AI governance is evolving to address these ethical challenges directly. Laws are beginning to shift from focusing solely on technical concerns to addressing broader ethical questions.
First of all, the improvement of ethical reviews is a way to boost the healthy development of DS technology, which is one of DS service providers’ obligations. Pursuant to the DS Administrative provisions, service providers should demonstrate the value, morality and safety of their DS technology in accordance with relevant laws and regulations as well as industrial standards.
In the meanwhile, relying on DS techniques, AI may be used to generate false information for illicit activities, as exemplified by prevalent face-swapping scams. The large-scale dissemination of AI-generated false information has the potential to not only infringe upon individuals’ legal rights, but also to worsen the online ecosystem and pose risks of misleading the public. For example, the abuse of AI face-swap technology, one the one hand, constitutes a universal threat to the rights related to personal dignity, such as reputation and image rights, and may further seriously infringe on people’s property right. On the other hand, such technology can also cause a crisis of trust in society, as the effectiveness of some civil actions will be disputed. For example, it is now somewhat difficult to prove that the person signing the contract is a real person, when the whole process happens on video links. To address these unintended harms, both legal requirements and ethical principles are reflected in the DS Administrative Provisions.On a separate note, in the development process of AI technology, challenges remain in terms of algorithm improvement, privacy protection and data security both domestically and globally. There were similar cases occurring in the United States, which involves OpenAI and Scarlet Johansson, alleging OpenAI engaged in misappropriation of her voice. The world has reached the consensus that technological innovation knows no bounds, but its application should be within legal and ethical boundaries. The application of DS technology, therefore, should adhere to the principle from both legal and ethical fronts, which has been raised in the evolving legal landscapes and further addressed in relevant Chinese laws, regulations, guides, standards, norms and so forth.
See Xinhua News Agency, AI ‘resurrection’ can’t be outside law’s purview, China Daily (April 9, 2024).
For further information, please contact:
Xiangxiang MA, Partner, Anjie Broad
maxiangxiang@anjielaw.com
- See article 3 of the DS Administrative Provisions.
- See article 4 of the DS Administrative Provisions.
- See article 7 of the DS Administrative Provisions.
- See article 8 of the DS Administrative Provisions.
- See paragraph 1 of article 6 of the DS Administrative Provisions.
- See article 9 of the DS Administrative Provisions.
- See paragraph 2 of article 6 of the DS Administrative Provisions.
- See article 10 of the DS Administrative Provisions.
- See article 11 of the DS Administrative Provisions.
- See article 12 of the DS Administrative Provisions.
- See 21ST CENTURY BUSINESS HERALD, Rise of DeepSeek sheds light on Chinese AI path, China Daily (February 10, 2025).
- See chapter III Data and Technology Management Standards of the DS Administrative Provisions.
- See article 14 of the DS Administrative Provisions.
- See article 15 of the DS Administrative Provisions.
- See Zhang Linghan, AI ‘resurrection’? Let bygones be bygones, China Daily (January 3, 2025).
- See supra note 14.
- See article 17 of the DS Administrative Provisions.
- See article 3 of the Draft Regulations on Labeling AIGC.
- See article 4 of the Draft Regulations on Labeling AIGC and article 17 of the DS Administrative Provisions.
- See article 6 of the Draft Regulations on Labeling AIGC.
- See Standard Practice Guide: Coding Rule for Service Providers of Network Security Artificial Intelligence Generated Synthetic Content Label (Draft for Comment), published on January 22, 2025.
- The Algorithms Administrative Provisions is jointly issued by CAC, MIIT, MPS and the State Administration for Market Regulation on December 31, 2021, taking effect on March 1, 2022.
- See article 4 of the Algorithms Administrative Provisions.
- See chapter II and chapter III of the Algorithms Administrative Provisions.
- See article 24 of the Algorithms Administrative Provisions.
- See article 19 of the DS Administrative Provisions.
- See supra note 4.
- See article 4 and article 5 of the DS Administrative Provisions.
- See article 4, article 7 and article 17 of the DS Administrative Provisions.
- See Todd Spangler, Scarlett Johansson Says She Was ‘Shocked’ and ‘Angered’ Over OpenAI’s Use of a Voice That Was ‘Eerily Similar to Mine’, Variety (May 20, 2024).
See supra note 1.
