About the legal regulation of generative artificial intelligence in China

Sayapin Sergei Petrovich ORCID: 0009-0005-3296-9929 PhD in Law Junior Researcher; Civil and Business Law Sector; Institute of State and Law of the Russian Academy of Sciences 117535, Russia, city, Moscow, passage. 3rd Dorozhny, 8, sq. 1 spsayapin@yandex.ru

Technologies and mechanisms of generative artificial intelligence (hereinafter also abbreviated) are used to create new (often quasi-new) data (text, audio, video, photo and other materials). In the field of medicine, pharmaceutical companies use GII to design formulas for new medicines, and in finance, GII is used to identify and predict market trends. All this has become possible mainly due to the achievements of modern science in the field of neural network architecture and information technology.

The principle of operation of GII is the "self-learning" of an information system based on large amounts of input data. The GII model studies static and dynamic patterns, as well as the characteristics of input data. For example, GII text models work on huge amounts of text data, revealing the structure of the language, the style of writing the text, the context of the use of individual lexemes, etc., then, GII can generate meaningful and coherent text based on the information received and processed.

Due to the rapid development of information technologies and technical means, as well as a wide range of applications of generative artificial intelligence ^{[6, p. 73][7]}, investments in this area are increasing. Since the early 2010s, there has been a significant increase in venture capital investments in start-ups engaged in the development of GII. Companies like OpenAI and DeepMind have already raised tens of billions of dollars in venture capital to develop generative artificial intelligence. Large technology companies are actively investing in research and development of GII, and the emergence of digital platforms such as SaaS (English: Software as a Service), providing access to GII through cloud services, make these technologies more accessible to a wide range of users, which only stimulates further development of this area. It should be noted that the governments of many leading countries of the world are also beginning to realize the scale of GII's capabilities and invest in the development of these technologies for national security and the promotion of state interests at the regional and international levels. According to published information, China has been actively developing the use of neural networks in shaping public opinion for several years, involving its large technology companies Midu, Suishi, Goertek, and others^[10].

The development of generative artificial intelligence technologies has become possible, first of all, "due" to the absence of strict regulatory frameworks, however, new forms of information technology always generate new risks for society and state security: the provision of false and misleading information, the spread of fraudulent schemes of money withdrawal, other types of Internet threats, therefore, the expansion of The use of GII has a reverse (in this sense, a negative) side. Since approaches to reducing such risks have not yet been fully developed in the world, each state independently seeks a balance in the regulatory and legal regulation of AI (GII). Therefore, discussions are increasingly taking place about such rules of operation of the GII, which will allow not only to maintain favorable conditions for the development of this technology and attract investment, but also to ensure state security.

Today, there are mainly two main approaches to the formation of domestic policy regarding the legal regulation of activities related to artificial intelligence (including generative AI).:

· The first approach is to develop and adopt general rules and regulations (most often declarative) regarding artificial intelligence, without adopting special laws regulating AI. This is exactly the way the United States is currently going, providing freedom for the development of AI in the country. So, most recently, in January 2025, a presidential decree was signed in the United States "On removing obstacles to American leadership in the field of artificial intelligence," according to which the policy of US global dominance in the field of AI is emphasized. The federal executive authorities of the United States of America have been tasked with developing a national action plan for the development of artificial intelligence ^[11]. There are, of course, separate legislative initiatives at the US state level, however, there is no need to talk about AI regulation in the US now.;

· The second approach is related to the development of special regulatory legal acts on the regulation of artificial intelligence (China, the European Union).

I must say that the goal of developing and developing artificial intelligence in China, as well as in the USA, is indicated as a priority ^{[1, p. 225]}. Therefore, back in 2017, the PRC adopted a Plan for the development of a new generation of artificial intelligence ^[12], but at that time, AI regulation, of course, did not yet exist there^{[4, p. 54]}. In the period from 2017 to 2022, China was only preparing for the legislative regulation of AI. The specified period of time was marked by the adoption of the White Paper on Standardization of Artificial Intelligence, as well as the Code of Ethics for Artificial Intelligence of a new Generation ^[8]. By the beginning of 2022, the Chinese Cyberspace Administration (CAC) began to develop (together with other government agencies) the first special regulatory legal acts to regulate activities using AI. According to some experts, China has since begun to pursue so-called "neutral regulation", due to the observance of the principle of equivalence between innovation, security and manageability in the development of regulatory legal acts on AI ^{[3, p. 255]}. However, in our opinion, it is hardly possible to call such a mechanism neutral regulation, due to the fact that, due to its specifics, there is a strong administrative component in the PRC, and the priority, first of all, of state ideology and national security – in other words, the priority of the public over the private ^{[2, p. 158]}, which is striking It is reflected, as a rule, in all legal acts of the People's Republic of China.

Speaking of China's special regulatory legal acts on AI regulation, the following should be briefly considered.

Thus, on March 1, 2022, the Regulation on the Management of algorithmic recommendations of Internet information Services came into force ^[13]. The document obliges persons using recommendation technologies to provide relevant services (recommendation technology service providers) to comply with the principles of fairness, openness and transparency in their work (art. 4). It is prohibited to use Internet services based on recommendation algorithms in order to create threats to the public interests and national security of the People's Republic of China (art.1).

The Regulation establishes state supervision over the providers of these services, as well as introduces public administration in this area (art. 3). All providers of algorithmic recommendation technology services are required to register with the state. The State maintains a register of such suppliers (art. 24).

The Regulation establishes a normative definition of the concept of "application of algorithmic recommendation technology" – the use of algorithmic technologies such as generation and synthesis, sorting, selection, search, filtering, personalized delivery, as well as decision-making planning to provide information to users (art. 2). According to art. 16, the service provider of algorithmic recommendation technologies is required to notify users about the goals, principles and mechanism of the provision of relevant services.

Service providers of algorithmic recommendation technologies are required to process requests from users of their Internet services within the prescribed time limit (art. 22).

At the same time, it should be noted that in the text of the Regulation itself there are no legally established definitions of such concepts as supplier, user or service provider of algorithmic recommendation technologies, etc. The meanings of these terms are derived from the literal interpretation of the Regulations.

The Regulation does not contain specific rules on the supplier's service agreement.

Chapter 5 of the Regulations establishes the responsibility of the service provider of algorithmic recommendation technologies (Articles 31, 32, 33). This is mainly administrative liability in the form of a warning or a fine in the amount of 10 to 100 thousand yuan, as well as, in some cases, criminal liability.

On January 10, 2023, the Regulation "On the Procedure for Managing Internet Information Services using Deep Synthesis Technologies" came into force ^[14]. The document is aimed at eliminating the risks associated with the use of deep synthesis technologies that allow deep modification of data for illegal purposes.

The Regulation (art. 23) establishes normative definitions of such concepts as:

· deep synthesis technologies are technologies that use deep learning, virtual reality and other algorithms for the generative synthesis of text, photo, video and audio materials and other information, including methods of working with these materials (see paragraphs 1-6 definitions of this concept).

· Deep synthesis service providers are organizations and individuals providing deep synthesis services;

· Deep Synthesis technical support service providers are organizations and individuals that provide deep synthesis technical support services;

· Users of deep synthesis services are organizations and individuals who use deep synthesis services for the purpose of producing, reproducing, publishing, or distributing information;

· Training data is an annotated or reference dataset that is used to train AI models.

The provision of deep synthesis services must comply with the principles of legality and morality, as well as the public interests and national security principles of the People's Republic of China (art. 4). Deep synthesis services must not be used to commit illegal acts and publish false and misleading information (art.6). Deep synthesis service providers are required to protect the received data, counter fraud in the process of deep synthesis of data, as well as in the process of storing, processing, using and transmitting the received data (art. 7), develop, disclose and periodically improve the rules of the platform where deep synthesis services are provided (art. 8), identify users of deep synthesis services (art.Article 9), analyze the received information (input data) and generated data for false and illegal information, take measures to restrict access to it and eliminate it (Article 10). Internet resources (including platforms) on which deep synthesis services are provided should provide for the possibility of processing user requests (art. 12).

If text, photos, audio, video, and other types of data are generated by AI, they must be marked with special markings, and the deep synthesis service provider must keep a log of the generation of AI materials (art.16).

The Deep Synthesis Regulation itself does not explicitly establish any liability measures for non-compliance with the norms set out in it. Instead, Article 22 states only that appropriate sanctions may be applied in accordance with other applicable Chinese regulations if any deep synthesis service provider violates the provisions of the Regulation. It also does not specify the consequences of non-compliance with the rules by users of deep synthesis services or digital platforms using which deep synthesis services are provided.

Nevertheless, the Regulation gives cyberspace departments certain powers to monitor compliance and conduct inspections. Cyberspace departments and other government agencies in China have the right to oblige suppliers to eliminate violations if, during the audit, it is determined that deep synthesis services pose significant information security risks to the People's Republic of China.

On August 15, 2023, Temporary Measures for the management of generative artificial intelligence services came into force ^[15]. The document establishes the principles according to which the provision and use of generative AI services must comply with the laws and administrative regulations of China, comply with socialist values, materials created by AI should not harm China's state security, promote terrorism and separatism, etc. It is important to note that the document does not apply to the activities of state and educational institutions of the PRC that use GII in scientific research (art.

The document defines concepts such as:

· GII technologies are models and related technologies that have the ability to generate data (text, photo, video, audio, etc. types of materials).

· GII service providers are organizations and individuals who use GII technologies to provide GII services (including the provision of GII services for programmable interfaces, etc.);

· Users of GII services are organizations and individuals who use GII services to create text, video, photo, audio and other types of materials (content).

To create AI models and generate AI of new materials, AI service providers are required to use only legal data sources; while respecting intellectual property rights; if information related to personal data of individuals is used to generate data, such information should be obtained only with their consent; if the data, subsequently If they are used in training, then the GII service provider is obliged to "work to improve the reliability, accuracy and objectivity of the specified data" (art. 3).

Temporary measures for managing generative artificial intelligence services oblige the GII service provider to conclude a service agreement with each user of its service. The contract must specify: the subject, terms, rights and obligations of the parties when using the GII service, as well as other provisions (art. 9).

Temporary measures to manage generative artificial intelligence services provide for a person's right to access, modify, and delete information about themselves from the AI database. In the event that the GII service provider receives a corresponding request, it is obliged to process it in a timely manner (art. 11). However, the scientific literature notes that due to the lack of a specific deadline for reviewing requests, as well as taking into account the fact that requests are most often considered not by humans, but by artificial intelligence, the user usually falls into an "endless wait"^{[5, p. 60]}.

Temporary measures to manage generative artificial intelligence services oblige to combat illegal content not only by GII service providers, but also by communication providers. In the event that a communications provider discovers illegal materials, it is obliged to immediately terminate access to them by network users (art.14).

According to Article 21, if a service provider violates the Temporary Measures for Managing Generative Artificial Intelligence Services, then a warning may be issued against the supplier; or an order to eliminate violations; or penalties may be applied (however, the specific amount of the fine is not set); or the right to provide relevant services in China may be suspended.

If the regulatory acts of the People's Republic of China are violated in the process of providing AI services by non-resident suppliers of the People's Republic of China, the Chinese Cyberspace Administration has the right to apply to the relevant Chinese authorities with a request to take measures to block Chinese users' access to these service providers.

In 2023, changes to Measures for the Ethical Examination of Science and Technology were proposed for public discussion (public discussions ended on May 3, 2023) ^[16]. The document establishes the provisions that within the framework of universities and research organizations in China, special committees on scientific and technological ethics should be established, whose responsibility is to conduct an examination of scientific research and scientific and technical activities from the point of view of ethics and morality. Scientific research in the field of artificial intelligence (including AI) is also subject to expertise, among other areas.

In March 2024, the draft law "On Artificial Intelligence" was published in China ^[17]. The text of the document was proposed for general discussion by the Chinese scientific community.

Chapter I of the draft law proclaims the principle of developing the field of AI in combination with state control (supervision) of the industry (art. 1), as well as the principle of compliance of the AI industry with ethical measures adopted in China (art. 3). It should be said that the scope of the draft law extends not only to public relations using AI within China's borders, but also beyond its borders, in the event that legal relations affect the rights and legitimate interests of Chinese individuals and legal entities, as well as the national security of the People's Republic of China (art.

Based on the literal interpretation of the text of the norms of Chapter II, it can be concluded that the developers of the draft law formulated three key (one might even say, to some extent, ideal) aspects of the functioning of artificial intelligence:

· availability of public computing power;

· Availability of open source AI models and algorithms;

· and also, the availability of shared data resources with the possibility of sharing them.

According to the developers of the bill, the state, while respecting the main principle – China's national security, should develop and encourage these areas of AI development.

Chapter III establishes that the right to data should be considered as one of the key rights; the possibility of AI being a subject of law is unequivocally denied (only individuals and legal entities can be subjects of copyright and patent law); data obtained as a result of AI's work is protected by intellectual property law; data obtained as a result of AI's work, It can be recognized as an invention or a work; the bill gives the right, subject to certain conditions, to use data as input for AI without paying remuneration to its owner (the so-called "reasonable use of data").

According to the norms of Chapter IV, the law distinguishes between general AI and critical AI, emphasizing the special features inherent in critical AI in terms of the formal organization of the model, risks, emergency response, etc., therefore, special measures must be taken in each of the sectors of the economy to regulate critical AI.

Chapter V is devoted to state control (supervision) of activities using artificial intelligence. For more effective control (supervision) of activities using AI, the gradation (categorization) of artificial intelligence is being introduced. Chapter VI is devoted to the use of AI in special fields (government, medicine, justice, mass media, etc.). Chapter VII is devoted to international cooperation, and Chapter VIII establishes the types of legal responsibility.

To date, the draft law on AI proposed by Chinese legal scholars is still under discussion, but it is already clear that it significantly complements and expands the already established architecture of regulatory regulation of artificial intelligence in the People's Republic of China. It contains a lot of bold ideas and meanings (for example, about the legal protection of data obtained as a result of the work of the GII). It seems that during 2025, the specified draft law (apparently with improvements) will still be adopted.

Based on the existence of regulatory legal acts that have already entered into force and are currently in force regarding artificial intelligence (including generative), as well as trends towards the rapid formation of the basic law on AI, it clearly follows that China is following the path of regulatory regulation of this area for general use within the PRC, while giving freedom of use and study AI for government purposes, in order to protect national interests.

^[12] 国务院关于印发新一代人工智能发展规划的通知 [Notification of the State Council on the publication of a Plan for the development of a new generation of artificial intelligence]. - URL: https://www.gov.cn/zhengce/content/2017-07/20/content_5211996.htm (date of request: 03/01/2025).
^[13] 互联网信息服务算法推荐管理规定 [Regulation on the management of algorithmic recommendations of information Internet services]. - URL: https://www.cac.gov.cn/2022-01/04/c_1642894606364259.htm (date of request: 03/01/2025).
^[14] 互联网信息服务深度合成管理规定 [Regulation "On the procedure for managing Internet information Services using Deep Synthesis Technologies"]. - URL: https://www.cac.gov.cn/2022-12/11/c_1672221949354811.htm (date of request: 03/01/2025).
^[15] 生成式人工智能服务管理暂行办法 [Temporary measures for managing generative artificial intelligence services]. - URL: https://www.cac.gov.cn/2023-07/13/c_1690898327029107.htm (date of request: 03/01/2025).
^[16] 关于公开征求对《科技伦理审查办法（试行）》意见的公告（已结束) [announcement on the public gather views on the efforts to revise the ethics of science and technology]. - URL: https://www.most.gov.cn/wsdc/202304/t20230404_185388.html (date of request: 03.03.2025).
^[17][The Law on Artificial Intelligence (draft proposals of scientists)]. - URL: http://www.fxcxw.org.cn/dyna/content.php?id=26910 (date of request: 03.03.2025).