[author]Shen Weixing

[content]

Future-Oriented Artificial Intelligence Legislation in China: Approaches and Key Priorities

Author: Shen Weixing

Professor, Tsinghua University School of Law

Doctoral Supervisor

Abstract: AI governance is gradually transitioning from soft law to hard law. China's AI legislation has been explicitly incorporated into its legislative planning. Our current objective is to formulate a law that meets China's AI security governance and industrial development needs while aligning with the international consensus on AI governance. At this stage, the emergence of AI law is necessary for three main reasons: First, to safeguard the standardized and healthy development of AI. The widespread application of AI brings risks related to trust, fairness, loss of control, society, and liability, all of which require legal frameworks for resolution.

AI governance is gradually transitioning from soft law to hard law. China's AI legislation has been explicitly incorporated into its legislative planning. Our current objective is to formulate a law that meets China's AI security governance and industrial development needs while aligning with the international consensus on AI governance. At this stage, the emergence of AI law is necessary for three main reasons: First, to safeguard the standardized and healthy development of AI. The widespread application of AI brings risks related to trust, fairness, loss of control, society, and liability, all of which require legal frameworks for resolution. Second, to ensure the innovative development of the AI industry. With the maturation, application, and promotion of generative and discriminative AI technologies, an increasing number of institutional obstacles in AI technology and industrial development need to be addressed through legal systems to ensure the industry's progress. Third, to support China's participation in international AI governance as a responsible major power. Currently, Western countries, including Europe and the United States, are accelerating their AI legislative efforts, using pioneering AI laws to establish their discourse systems. This limits the space for international cooperation in China's AI technology R&D and will inevitably have a significant impact on the international market for Chinese AI products. In response to this trend, China must act proactively, vigorously promote AI legislation, and build the prerequisite and foundation for participating in global AI governance.

Feasibility of AI Legislation in China

First, the comprehensive development of the AI industry provides a social foundation for legislation. Currently, the performance of algorithms based on large-scale pre-trained models like BERT and GPT continues to improve. AI-generated content (AIGC) technology is widely used in social media, broadcasting, and television. Single-point computing power continues to increase, with customization and diversification of computing power becoming important development trends. AI can be deployed in more terminal products, facilitating the popularization of new technologies such as autonomous driving and telemedicine. Furthermore, AI technologies represented by deep learning universally use large amounts of labeled data to shape intelligent cognition. Precision marketing and automated decision-making technologies are being widely applied in e-commerce, internet finance, and public administration. The EU AI Act has already set norms and standards for constructing AI regulatory sandboxes. In November 2023, Spain was the first in the EU to establish a regulatory sandbox. Subsequently, at the 2024 Zhongguancun Forum, China launched its first domestic AI regulatory sandbox, with initial selected enterprises including ZKWG, TRS, and Lingxi Cloud. With the continuous development of AI technology, the implementation and application of related projects have shown an accelerating trend in recent years.

Second, a series of AI laws and regulations issued by central and local authorities has formed a working basis for legislation. In recent years, central and local authorities have formulated a batch of departmental rules, judicial interpretations, and local regulations concerning comprehensive AI legislation, legislation for specific technological directions, and legislation for specific industries, accumulating working experience for national legislative bodies in AI legislation. In terms of comprehensive AI legislation, the "Shenzhen Special Economic Zone Artificial Intelligence Industry Promotion Ordinance" and the "Shanghai Municipality Regulations on Promoting the Development of the Artificial Intelligence Industry" were promulgated successively in 2022, making local explorations with a legislative approach that prioritizes development while considering security governance. In terms of legislation for specific technological directions, the "Provisions on the Management of Algorithmic Recommendations in Internet Information Services" issued by the Cyberspace Administration of China (CAC) in December 2021 deeply explored algorithm technology governance and implemented filing and inspection for high-risk algorithms. The "Provisions on the Management of Deep Synthesis in Internet Information Services" issued in November 2022 explored the governance of AI technologies that use deep learning, virtual reality, and other generative synthesis algorithms to produce text, images, audio, video, virtual scenes, and other online information, and prospectively provided security assurance rules for the disruptive applications of generative AI technologies like ChatGPT. The "Interim Measures for the Management of Generative Artificial Intelligence Services" came into effect in August 2023, making China the first country to legislate specifically for generative large models. Additionally, relevant provisions issued by the Supreme People's Court regarding facial recognition technology provide a legal basis for its regulation. In terms of legislation for specific industries, supportive legislation by departments such as the Ministry of Transport and the Ministry of Natural Resources in fields like autonomous driving has provided a legal guarantee for the healthy development of AI. This work has laid a solid foundation for future AI legislation.

Third, AI legislation by international organizations and Euro-American countries provides international experience for reference. International organizations and Euro-American countries have respectively formed AI legislative working groups and have released a series of AI laws (including drafts) and preparatory legal documents. The EU adopted the "AI Liability Directive" in September 2022, establishing applicable rules for damages caused by AI systems. The EU's "AI Act," passed in 2024, proposes a risk-based regulatory approach. The United States released the "Blueprint for an AI Bill of Rights" in October 2022, aiming to help guide the design, use, and deployment of automated systems to protect civil rights in the AI era. The OECD adopted the "Recommendation on Artificial Intelligence" in May 2019, the World Health Organization adopted the "Ethics and Governance of Artificial Intelligence for Health" guidance in June 2021, and UNESCO adopted the "Recommendation on the Ethics of Artificial Intelligence" in November 2021. These documents are accompanied by supporting research reports and legislative direction explanations, which can serve as references for China's AI legislation.

Positioning and Direction of AI Legislation in China

In China's AI legislative work, it is crucial to identify the key and difficult issues in legislation and clarify their solutions.

First, the positioning of AI legislation should find an appropriate balance between being a promotional law and a regulatory law. AI legislation receives considerable attention, and a core issue is whether the law aims to strengthen regulation or promote development; this positioning is of great concern to all sectors of society. If a law is merely promotional, it resembles industrial policy, lacks legal weight, and its legislative value is difficult to demonstrate. If legislation overestimates AI risks and imposes excessive obligations, it may hinder the rapid development of AI in China, causing us to miss development opportunities in increasingly fierce international competition—this is the primary concern of the AI industry regarding legislation. Therefore, the stance and positioning of future AI law are the foremost macroeconomic questions to answer. In this regard, the author believes that security and development are not oppositional; lack of development is the greatest insecurity. At the same time, risky technological innovation must be sustainable under the premise of security. Domestic and international AI legislative and practical frontiers also indicate that establishing security safeguards in AI legislation helps enhance the trust of application departments and consumers in AI products. Unfettered leniency will only lead to the stigmatization of our AI products, ultimately causing technology to be stifled by security requirements. Research in the AI industry shows that the application and implementation of technology require AI industry promotion measures and the resolution of institutional constraints through legislation, providing legal support for science and technology policies.

Second, the form of AI legislation should comprehensively consider the systemic integrity of unified legislation and separate legislation. Regarding the form of AI legislation, China needs to consider whether to enact a comprehensive law or to legislate separately for different fields, industries, and scenarios. The author's answer is that AI legislation should comprehensively consider the systemic integrity of unified legislation and separate legislation. The specific approach is to prioritize comprehensive legislation, supplemented and supported by specific laws. Comprehensive legislation refers to a general law on AI formulated by the Standing Committee of the National People's Congress, uniformly legislating on the basic principles, general rules, management systems, operational mechanisms, comprehensive measures for promoting development, legal responsibilities, etc., related to AI technology development. Furthermore, such a comprehensive law cannot solve all problems and will require the drafting of supporting regulations for different AI application scenarios and life cycles, such as separate legislation for autonomous driving, precision medicine, and government use of AI for decision-making. This coordination of macro and micro perspectives can meet the needs of phased legislation and continuous improvement, thereby adapting to the characteristics of continuous iteration and evolution of AI technology. Considering the wide applicability and importance of AI technology, comprehensive legislation is particularly necessary. At the same time, to avoid conflicts in the legal system and market disruption, a strategy of central legislation first should be adopted to create a legal environment that encourages technological innovation.

Third, the value of AI legislation lies in addressing the pain points and bottlenecks in the development of the AI industry. Technological development is, in principle, the result of free market competition and does not require legal intervention; many "promotional laws" often become mere formalities and waste legislative resources. However, legal systems can also promote industrial development through rules with binding force. First, by delineating responsibility boundaries to enhance investment expectations, such as clarifying the responsibility boundaries of AI developers and operators, and adhering to the "safe harbor rule" in the AI field to promote industrial development. Second, by solving the problem of inefficient market resource allocation, establishing coordination mechanisms for data supply, computing power supply, and application deployment through legislation to promote the development of AI technology. Third, by building an industry ecosystem, providing support measures in fiscal support, talent cultivation, and industry incentives through national legislation, thereby promoting the formation of a specialized AI industry ecosystem.

Fourth, the operational space for AI law lies in managing the relationship between AI legislation and existing relevant laws. In the process of China's AI legislation, clarifying its relationship with existing relevant laws is crucial. This not only helps build a coordinated and unified legal system but also ensures the applicability and foresight of the law. First, in promoting scientific and technological development, AI legislation should fully draw upon and absorb the main content of the "Law on Scientific and Technological Progress." This will ensure that AI, as an important driving force for scientific and technological progress, can be effectively promoted and supported within the legal framework. Second, clarify the relationship between AI legislation and the "Cybersecurity Law," "Data Security Law," and "Personal Information Protection Law." Their connection lies in their shared involvement with data processing activities, jointly forming the legal guarantee system for the digital economy. The difference is that AI legislation focuses more on regulating the behavior of algorithm operation rules, concentrating on how AI systems automatically simulate, assist, or replace human behavior, and create new human-computer interaction relationships. In this process, data input and output processing still need to comply with the provisions of the "Personal Information Protection Law" and "Data Security Law," while algorithm rules themselves become the core of AI legislation. The development process of AI systems, including crucial links like model selection and parameter training, is vital to system performance but often exceeds the regulatory scope of traditional data processing laws. The focus of AI governance is the intelligent social order; the computational model structure based on algorithms in AI actually becomes a rule for a virtual society. Governing virtual space involves risks of opacity, unfairness, and insecurity. Its governance mechanism often does not aim to protect individual control rights but rather aims to control the rules of digital society or the digital social environment, or public order. Therefore, the new AI law should particularly emphasize the logic of public administrative supervision, and administrative organs need to play a greater role in protecting private rights. Simply put, the "Cybersecurity Law," "Data Security Law," and "Personal Information Protection Law" constitute the fundamental regulations for AI legislation, but they all lack the specificity and systemic integrity to regulate AI. AI legislation can be said to be an extension and specialization based on these three laws. Third, clarify the relationship between AI legislation and the "Provisions on the Management of Algorithmic Recommendations in Internet Information Services," the "Provisions on the Management of Deep Synthesis in Internet Information Services," and the "Interim Measures for the Management of Generative Artificial Intelligence Services."1 These three departmental rules are highly targeted and constitute the working basis for future AI legislation. Regarding the relationship between AI legislation and these three rules, the author's suggestion is that the important content of these three rules should be absorbed into the future AI law, and their specific detailed provisions can appear as supporting regulations for the future AI law.

Fifth, how to grasp the legislative pace for AI. Whether now is the best time for legislation has always been a point of hesitation. There are calls from the AI industry to "let the bullets fly a bit longer." However, we must clearly recognize that the risks brought by AI development, such as the erosion of self-determination and the generalization of high-risk technologies, require legislation to ensure its development on a safe track. At the same time, breaking down "data silos" and establishing systems for AI development and rational data use also need to be resolved through legislation. Currently, China urgently needs to provide traction for AI in both development and security dimensions to avoid unfavorable situations in international competition and systemic risks. Given that the EU and the US have already initiated comprehensive AI legislation, China also needs to have relevant stances and measures to demonstrate the safety and reliability of its AI products, ensure product transparency and security, and enable both technology and law to enter the international market together. Considering the long legislative cycle and the lack of consensus among all parties on the direction of AI legislation, it is recommended that the state propose a draft AI law as soon as possible for full discussion by all sectors of society, and pass China's "Artificial Intelligence Law" when the time is ripe.

Key Issues in AI Legislation in China

China's AI legislation should adopt a legislative system that combines comprehensive legislation with industry-specific supporting legislation to meet the needs of China's AI governance for fundamental national legal systems through legislative forms. In the process of China's AI legislation, establishing a series of basic principles and specific systems is crucial. The construction of this system should not only promote the healthy development of AI and protect public interests but also maintain coordination with international standards and practices, ensuring China's competitiveness and voice in the global AI field.

First, the legal definition of AI and its scope of regulation. The definition of AI should cover automated systems from perception to decision-making, including machine learning, logic, and knowledge-based systems, and define its characteristics. The scope of regulation should cover the rights and obligations of developers, operators, and users, as well as the geographical scope of domestic and foreign systems, ensuring the universal applicability and specificity of the law. Second, the basic principles of AI law. Legislation should establish four basic principles: encouraging innovation, collaborative governance, public participation, and security governance, aiming to promote technological development, ensure the participation and supervision of all social parties, and build public trust and confidence in technology. Third, establish an AI risk management system. Drawing on the experience of the EU AI Act, China's future "Artificial Intelligence Law" should establish an AI risk assessment and classified/graded governance system, scientifically dividing application categories and risk levels. Based on the degree of human involvement and the impact on human rights, interests, and social order, risks should be divided into four categories: unacceptable, high, limited, and low, and corresponding regulatory rules and assessment methods should be established. The establishment of this system helps to adopt differentiated management measures for different types of AI applications, ensuring a balance between risk management and innovative development. Fourth, improve regulatory measures such as "transparency" and filing. Establish transparency and filing requirements based on risk classification, clarifying the subjects, content, and procedures for filing. Establish risk management systems for high-risk systems, require transparency and user's right to know for limited-risk systems, and encourage the development of codes of conduct for low-risk systems. Fifth, improve mechanisms for AI damage relief and burden sharing. To enhance victims' ability to obtain relief, legislation needs to clarify the principles of tort liability and establish a liability insurance system. Liable parties may include suppliers, importers, distributors, and users, determined specifically based on control and risk disposal capabilities. High-risk systems must purchase liability insurance to ensure that victims receive timely and adequate relief in the event of damage. In a modern risk society, victim relief cannot solely rely on legal liability mechanisms; an AI development damage relief fund should also be established and improved to disperse damage risks, alleviate the pressure on AI industry development, and promote AI innovation. Sixth, improve collaborative supervision and corporate/industry self-regulation mechanisms. Implement the administrative reform concept of streamlining administration, delegating power, improving regulation, and optimizing services; avoid regulatory overlap and vacuums; establish a regulatory mechanism coordinated by the central Cyberspace Administration with a division of labor among multiple departments; and leverage the supervisory role of industry organizations and the public. Through technical standards, risk assessments, and other measures, strengthen the autonomous role of industry organizations, encourage public participation and supervision in the R&D and operation of AI, and form a new pattern of whole-of-society co-governance.

Historically, disruptive technological innovations have always led to new legislation. New technologies represented by general artificial intelligence, as a globally recognized disruptive technological innovation, will inevitably lead to specialized legislation. The General Office of the State Council, in its 2023 legislative work plan, first proposed "preparing to submit a draft Artificial Intelligence Law to the Standing Committee of the National People's Congress for deliberation." The legal name in this plan indicates that China's AI legislation has chosen a comprehensive legislative route, which can simultaneously encompass the dual legislative goals of security governance and industrial development. The "Legislative Plan of the 14th National People's Congress Standing Committee" also proposed in its Category I legislative plan: "Promote scientific and technological innovation and the healthy development of artificial intelligence... requiring the formulation, revision, abolition, or interpretation of relevant laws, or decisions to be made by the National People's Congress and its Standing Committee, to be scheduled for deliberation at an appropriate time." This plan considers both the urgency and flexibility of AI legislation, which is conducive to relevant departments actively advancing related work. Therefore, we must have a composed mindset and a sense of urgency ("time waits for no one") to promote the promulgation of China's "Artificial Intelligence Law."