Interim Measures of Generative AI Services Released

Authors: Kevin DUAN丨Kemeng CAI丨Jin JIN

On July 13, 2023, the Cyberspace Administration of China ("CAC"), together with the National Development and Reform Commission, the Ministry of Education, the Ministry of Science and Technology, the Ministry of Industry and Information Technology, the Ministry of Public Security, and the National Radio and Television Administration issued the Interim Measures for Generative Artificial Intelligence Services (the "Interim Measures"), which will come into effect on August 15, 2023.  The Interim Measures have been substantially revised based on public feedback received in response to the Measures for Administration of Generative Artificial Intelligence Services (the "Draft"), which the CAC issued for public comments on April 11.  Compared to the Draft, the Interim Measures provide greater tolerance for immature generative AI services and place more emphasis on encouraging technological development and innovation, while better balancing development and security at the principle and institutional levels.  In this article, we briefly analyze the Interim Measures in terms of the scope of application, regulatory approach, interplay with the existing system, data training, service application, and foreign investment restrictions, with a focus on the optimizations and innovations in the Interim Measures based on the Draft, as well as their potential impacts.

The applicable scope emphasizes the provision of services to the domestic public, excludes research and internal applications

Article 2 of the Interim Measures stipulates that "these Measures shall apply to the utilization of generative artificial intelligence technologies to provide services for generating text, images, audio, video, and other content to the public within the territory of the People's Republic of China (hereinafter referred to as GAI services)", emphasizing that the target of the Interim Measures is the provision of GAI services to the public in the People's Republic of China.  The Interim Measures emphasize that the target of regulation is the provision of GAI services "to the public within the territory of the People's Republic of China".  Compared to the Draft, the Interim Measures incorporate public comments and exclude from their scope of application the research, development and application of GAI technologies that do not provide services to the domestic public.  This greatly reduces the compliance burden at the model development stage, while the latter alleviates the compliance concerns of many enterprises when accessing GAI services for internal application purposes such as improving work efficiency, reflecting the Interim Measures' prudent and inclusive regulatory approach and the policy goal of encouraging innovation.

Inclusive and prudent, categorized and classified regulatory approach, emphasizing the coordination of multisectoral regulations

The Interim Measures call for an "inclusive and prudent, categorized and classified" regulation of GAI services.  The Interim Measures add the Law of the People's Republic of China on Scientific and Technological Progress as their legislative basis, which highlights the policy orientation of encouraging scientific and technological innovation.  In addition, the Interim Measures add the National Development and Reform Commission, the Ministry of Education, the Ministry of Science and Technology, the Ministry of Industry and Information Technology, the Ministry of Public Security, and the National Radio and Television Administration as the relevant regulatory authorities, stipulating that each department will strengthen the management of GAI services in accordance with its own authority.

The classified approach to regulation may be inspired by the European Union's (draft) Artificial Intelligence Act, which categorizes AI systems into unacceptable risk, high risk, limited risk and minimal risk.  Due to the generic nature of GAI, the "inclusive and prudent, categorized and graded" regulatory approach will help the Interim Measures retain a certain degree of flexibility as the "basic law" in the field of GAI, on the basis of which regulatory authorities, industry authorities and standardization organizations can also formulate more detailed regulations for GAI.  Regulators, industry authorities, and standardization organizations can also formulate more detailed rules for the classification of GAI on this basis and set more stringent specifications for certain industries, application scenarios, and certain high-risk GAI services.  In addition, the Interim Measures address some of the major application scenarios of GAI services, stipulating that the use of GAI services to engage in activities such as news publishing, film and television production, and artistic and literary creation is subject to regulations in the relevant fields, which conform to the existing regulatory framework (Article 2).

Policy measures addressing practical issues and favoring innovation in the development of generative AI in the industry

Against the backdrop of the U.S. "decoupling" and "cracking down" on China's technology and the many obstacles to Chinese enterprises' access to advanced chips and computing capacity, the Interim Measures put forward a series of policy incentives for GAI research and development and application, including:

• Encouraging the innovative application of GAI technology in various industries and fields, generating positive and healthy, upwardly mobile quality content, exploring optimized application scenarios and building an application ecosystem.

• Supporting collaboration among industry organizations, enterprises, educational and scientific research institutions, public cultural institutions and relevant professional institutions in GAI technology innovation, data resource construction, transformation and application, and risk prevention.

• Encouraging independent innovation of basic technologies such as GAI algorithms, frameworks, chips and supporting software platforms, carrying out international exchanges and cooperation on an equal and mutually beneficial basis, and participating in the formulation of international rules related to GAI.

• Promoting the construction of GAI infrastructure and public training data resource platforms.  Promoting the categorized, graded and orderly opening of public data, and expanding high-quality public training data resources.  Promoting collaborative sharing of computing resources and enhancing the effectiveness of arithmetic resource utilization (Articles 5 and 6).

At present, some regions are already at the forefront of the construction and coordination of computing power, data and other infrastructure.  For example, the Implementation Plan for Beijing Municipality to Accelerate the Construction of a Source of Artificial Intelligence Innovation with Global Influence (2023-2025) released in May 2023 puts forward specific favorable policies in areas such as "enhancing the ability to coordinate the supply of computing resources", "strengthening the open sharing of public data", and "building an efficient and collaborative big model technology industry ecosystem".

Moderately loosening specific regulations on data training, model application, and optimization

The Interim Measures have drawn on feedback from the industry and more fully considered the current technical barriers and limitations in the quality of training data, reliability and accuracy of generated content, and moderately relax the specific compliance requirements for GAI data training, model application and optimization, which are mainly reflected in the following aspects.

• The Interim Measures appropriately relax the requirements on the quality of training data.  Article 7 replaces "ensure" the "authenticity, accuracy, objectivity, and diversity" of data in the Draft with "take effective measures to improve authenticity, accuracy, objectivity, and diversity", which reduces the responsibility of service providers in terms of the quality of training data.

• Article 4 of the Draft is deleted, which would have required that "the content generated by GAI shall be true and accurate, and measures shall be taken to prevent the generation of false information" and shall not contain "content that may disturb the economic order and social order".  This is replaced with a requirement that the provider "based on the characteristics of the service type, take effective measures to enhance the transparency of GAI services and improve the accuracy and reliability of generated content", which to a certain extent reduces the responsibility of the service provider in generating content.

• The requirement for real identity verification in Article 9 of the Draft is removed.  This may be due to the fact that current GAI services are mainly chat rather than publication, and many GAI services provided through programmable interfaces (APIs) may not be able to implement real identity verification obligations.  However, on the other hand, if GAI is used to provide internet information services, it may still be necessary to fulfill real-name identity authentication obligations in accordance with relevant regulatory requirements.

• Article 11 of the Interim Measures removes the requirement not to create user profiles and replaces the prohibition of providing user input information to others with a prohibition on collecting non-essential information and "unlawfully" providing user input information to others, i.e., excluding cases in which the user has given consent or in which there are other provisions in laws and regulations.  This amendment is more in line with the principles of informed consent and necessity of the Personal Information Protection Act and leaves more room for service providers to use user input information to optimize models and improve service quality.

• The Interim Measures reduce the obligation to monitor and dispose of illegal content by deleting the requirement in Article 13 of the Draft that providers should take measures to stop the generation of illegal content when they find or know of such content, by relaxing the requirement to establish a mechanism for the reporting of complaints and to deal with illegal information in a timely manner, and to rectify by taking measures such as optimizing model training and other measures.  The Interim Measures also did not adopt the controversial time limit of three months set in Article 13 of the Draft for preventing the generation of illegal content again through model optimization and training, leaving a certain amount of flexibility.

• The Interim Measures reduce the obligations of service providers in terms of algorithmic transparency by deleting the detailed requirement in Article 17 of the Draft for providers to "provide necessary information that may affect the trust and choice of users", and only requiring providers to "take effective measures to improve algorithmic transparency" in Article 4(5), allowing service providers to explore more flexible ways to improve algorithmic transparency.  Service providers can explore more flexible ways to improve the transparency of algorithms.

The Draft would have required that a security assessment of internet information services with public opinion attributes or social mobilization capabilities be conducted prior to providing services to the public using GAI products.  Article 17 of the Interim Measures clarifies that the subject of the security assessment obligation is the provider of GAI services that have "public opinion attributes or social mobilization capabilities", appropriately restricting the scope of the security assessment and maintaining consistency with the existing norms.

Foreign investment access and overseas services

Article 20 of the Interim Measures stipulates that if services provided by sources outside the PRC to the domestic territory do not comply with laws, administrative regulations, or the provisions of these measures, the national cyberspace department will notify relevant agencies to take technical measures and other necessary measures for disposal.  Combined with the provisions of Article 2 of the Interim Measures on the scope of application, we tend to believe that this article mainly aims to restrict foreign GAI services that do not comply with the Interim Measures from providing services to the domestic public.  In this case, the competent authorities may take technical measures such as blocking to cut off access to relevant overseas service websites and applications.  In addition, if domestic service providers embed foreign GAI services into their products to provide services to the domestic public, they need to comply with the relevant provisions of the Interim Measures; otherwise, the competent authorities may punish such domestic service providers in accordance with Article 21 of the Interim Measures.

Article 23 of the Interim Measures adds a provision that "if laws and administrative regulations stipulate that relevant administrative licenses must be obtained for providing GAI services, providers shall obtain licenses according to the law.  Foreign investment in GAI services shall comply with relevant laws and administrative regulations on foreign investment."  At present, laws and regulations have not set administrative licenses or foreign investment access restrictions on the provision of GAI services themselves, but GAI services will need to comply with relevant licenses or market access regulations if they are used in areas where licenses or foreign investment access exist, such as providing value-added telecommunications services, network audio-visual program services, and internet cultural operations.

Impact and outlook

Overall, the Interim Measures reflect many comments from the industry and the public on the Draft by taking into account the current technical limitations of generative artificial intelligence to a greater extent. The Interim Measures take a prudent and inclusive approach and moderately relax the compliance requirements for generative artificial intelligence from research and development, model training to application, fine tuning, and other stages.  The Interim Measures show an encouraging new technology development and application policy orientation.  However, looking at the specific rules, the Interim Measures requirements for training data compliance, generation content security and accuracy, transparency, etc., require enterprises to propose creative solutions to combine technology and legal compliance to alleviate the security concerns of regulatory agencies and win more institutional space for industrial development.