On July 13, 2023, the Chinese government published its finalized rules on generative artificial intelligence, the Interim Measures for the Management of Generative Artificial Intelligence Services ("Interim GAI Measures"), which came into effect on August 15, 2023 (Art. 24).1 The Interim GAI Measures were jointly approved by seven major Chinese government agencies, including the Ministry of Science and Technology, the Ministry of Public Security, and the Cyberspace Administration of China ("CAC"), Beijing's regulatory body for Internet matters. A draft version of the Interim GAI Measures was published in April 2023 to solicit public feedback.2
The objective of the Interim GAI Measures is to regulate generative AI, which is primarily designed to generate content, and they are the latest addition to an emerging framework of AI regulation in the People's Republic of China that already includes a number of application-specific and local laws. Chinese authorities introduced the Interim GAI Measures as regulators and policy leaders around the world grapple with the introduction of powerful generative AI tools such as ChatGPT and increasingly view their regulation as a policy priority. For instance, Group of Seven (G7) leaders at the G7 Summit in May jointly called for 'guardrails' on generative AI technology.3
China is emerging as a major player in this industry. According to some projections, the Chinese generative AI market is expected to grow from 66.3 billion renminbi (RMB) in 2022 to 207 billion RMB in 2025 (approximately C$12.8 billion to C$40 billion, respectively).4 In this context, Beijing's legislation on generative AI could potentially have significant implications for the AI industry in China and beyond.
Background of AI Regulation in China
The Chinese government started to support its AI industry at the national level early on. Beijing's 13th Five-Year Plan (2016-2020) specified AI as key for achieving economic growth targets,5 and the 14th Five-Year Plan (2021-2025) also signalled continued state investment in AI.6 On July 20, 2017, the Chinese government introduced its vision for the development of AI in China when it published its Next Generation Artificial Intelligence Development Plan.7 The Plan presented Beijing's comprehensive strategy for centering AI in China's efforts for socio-economic development and creating a trillion RMB (approximately C$200 billion) AI industry that would make China the world leader in AI by 2030.
Over time, a patchwork of AI laws came together at various levels in China under these broader strategies. At the regional level, China's first provincial law on AI development took effect on October 1, 2020 with the passage of the Shanghai Regulations on Promoting the Development of the AI Industry, which seeks to promote AI industry at the municipal level in Shanghai. Not long after, the Shenzhen government passed a similar law, the Regulations on Promoting Artificial Intelligence Industry in Shenzhen Special Economic Zone, which took effect on November 1, 2022.
Beijing also started to lay the foundation for specifically addressing generative AI systems. The Internet Information Service Algorithmic Recommendation Management Provisions ("Algorithmic Recommendation Provisions"), which set up the governance framework for regulating recommendation systems, came into effect on March 1, 2022.8 In addition, Beijing released more targeted rules for generative AI through the Internet Service Deep Synthesis Management Provisions("Deep Synthesis Provisions"), which came into effect on January 10, 2023 and apply to deepfake outputs from AI technology.9 The Interim GAI Measures build on the Deep Synthesis Provisions by providing more concrete rules and enforcement measures for generative AI specifically. Together with the Algorithmic Recommendation Provisionsand the Deep Synthesis Provisions, the Interim GAI Measures constitute the main legal basis for regulatory compliance and supervision of the AI industry in China.
Article 1 sets out the main purpose of the Interim GAI Measures, which is to "promote the healthy development and standardized application of generative artificial intelligence, safeguard national security and social public interests, and protect the legitimate rights and interests of citizens, legal persons and other organizations" in accordance with Chinese laws such as the Personal Information Protection Law.10 The Interim GAI Measures reflect Beijing's efforts to address the societal ramifications of generative AI, which is similarly reflected in the measures' subsequent emphasis on aligning the use of generative AI with "social morality and ethics" (Art. 4).
Article 3 highlights another organizing theme by placing "qual importance" on a range of priorities that include "development and security", "law-based governance", and "promoting innovation".
Scope and Key Terms
The Interim GAI Measures are distinct from the other aforementioned laws by specifically regulating the use of generative AI technology, defined as "models and related technologies that have the ability to generate content such as text, pictures, audio, and video" (Art. 22), to provide services for generating content to the public in the mainland of People's Republic of China (Art. 2). Compared to the Deep Synthesis Provisions, generative AI technologies subject to the Interim GAI Measures encompass more than algorithm-based generative and synthesizing technologies by also including models and rules-based systems.
The Interim GAI Measures apply to "providers" of generative AI services, defined as "organizations and individuals that use generative [AI] technology to provide generative [AI] services", including providing these services through application programming interfaces (APIs) (Art. 22(2)). As also indicated by "users" of generative AI services being defined as "organizations and individuals who use generative [AI] services to generate content" (Art. 22(3)), the Interim GAI Measures appear to encompass the provision of generative AI services indirectly to the public through business-to-business arrangements. However, institutions that develop and apply generative AI technology, but do not provide generative AI services to the public, are exempt (Art. 2).
In addition, the Interim GAI Measures appear to establish an extraterritorial scope by specifying that they apply "to the provision of services to the public within the mainland territory of the People's Republic of China" (Art. 2), potentially extending their application to individuals and organizations outside China that provide generative AI services to persons in the People's Republic of China. This is further nuanced by another provision stating that non-compliance with the Interim GAI Measures and other laws by generative AI providers outside the People's Republic of China will lead to notification of "relevant institutions to take technical measures and other necessary measures to deal with them" (Art. 20).
Notably, the Interim GAI Measures appear to have a significantly reduced scope compared to the draft version, which would have also applied to the "research, development, and use of products with generative AI functions" (formerly Art. 2). However, this wording is not referenced in the Interim GAI Measures.
General Requirements and Prohibitions
The Interim GAI Measures set out a number of general requirements for the provision and use of generative AI services(Art. 4):
- Respect for China's "social morality and ethics" and upholding of "Core Socialist Values" (Art. 4(1))
- "Effective measures" must be engaged during algorithmic design, training data selection, model generation and optimization, service provision, and other processes to prevent discrimination based on factors such as race, ethnicity, religious belief, nationality, region, sex, age, profession, or health (Art. 4(2))
- Respect for intellectual property rights, commercial ethics, and the protection of commercial secrets, and prohibition against uses for monopolies and unfair competition (Art. 4(3))
- Respect for the lawful rights and interests of others, and prohibition against endangering the physical and psychological wellbeing of others or infringing their rights and interests, including their image, reputation, honour, privacy, and personal information (Art. 4(5))
- Effective measures must be used to increase the transparency, accuracy, and reliability of generative AI services (Art. 4(5))
In addition, the provision and use of generative AI services must not generate content that could lead to the following outcomes (Art. 4(1)):
- inciting subversion of national sovereignty or the overturn of the socialist system;
- endangering national security and interests or harming the Chinese nation's image;
- inciting separatism or undermining national unity and social stability;
- advocating terrorism or extremism; or
- promoting ethnic hatred and ethnic discrimination, violence and obscenity, or fake and harmful information.
The Interim GAI Measures have more specific operational requirements for providers of generative AI services than the draft version. These operational requirements relate to a variety of matters such as model development, data governance, service standards, and content moderation, and include the following:
- Training Data: Providers must use data and foundational models with "lawful sources" (Art. 7(1)) and use "effective measures" to improve the quality, truth, accuracy, objectivity, and diversity of training data (Art. 7(4))
- Privacy Rights: Providers are responsible as handlers of personal information (Art. 9) and must obtain consent when using personal information unless an exception applies (Art. 7(3)) and comply with privacy rules including on data collection and minimization, data retention, and individual rights regarding access, correction, and deletion (Art. 11)
- Data Tagging: Providers must establish clear, specific, and practical tagging rules that meet the requirements of the Interim GAI Measures during the research and development process for generative AI (Art. 8)
- Content Moderation and Labelling: Providers are responsible as producers of online information content (Art. 9) and must label generated content as required by the Deep Synthesis Provisions (Art. 12), promptly address "illegal content" through "effective measures" such as stopping the generation or transmission of the content, removing it, and correcting it through model optimization training, and report the issue to the relevant authorities (Art. 14)
- User Engagement and Complaints: Providers must enter service agreements with users to clarify rights and obligations (Art. 9), use "effective measures" to prevent addiction or overreliance by minor users (Art. 10), and develop mechanisms for handling user complaints (Art. 15)
- Security Assessment: Providers of generative AI services with "public opinion properties"11or the "capacity for social mobilization" must conduct security assessments and comply with algorithmic filing requirements in accordance with the Algorithmic Recommendation Provisions (Art. 17)
Enforcement and Penalties
For violations, penalties are to be given by relevant regulatory departments in accordance with applicable laws and administrative regulations including the Cybersecurity Law of the People's Republic of China; Data Security Law of the People's Republic of China; Personal Information Protection Law of the People's Republic of China; and Laws of the People's Republic of China on Progress of Science and Technology, and where the violation does not address a particular area, applicable regulators may issue warnings, publish criticisms, and order providers to comply with the Interim GAI Measures. In "serious circumstances" or where such orders are refused, authorities may order the suspension of the generative AI services (Art. 21). However, in a notable change from the draft version, which would have authorized monetary penalties between 10,000 RMB and 100,000 RMB (approximately C$1,922 and C$19,220 respectively) for violations, the Interim GAI Measures do not prescribe monetary penalties.
Providers must also be prepared to cooperate with regulatory inspections by being able to explain the training data, including its sources, models, types, tagging rules, and algorithms, and providing any other necessary assistance (Art. 19). Authorities involved in the regulation of generative AI are also obligated to maintain the confidentiality of sensitive data such as commercial secrets and personal information acquired in the course of performing their duties, and must not disclose or unlawfully share this data with others (Art. 19).
Takeaways for Canadian Businesses and Organizations
Canadian businesses and organizations engaged in the generative AI industry and potentially serving customers in China should be aware of the requirements imposed by the Interim GAI Measures, as well as other regulations and guidance that form part of China's AI regulatory framework. The Interim GAI Measures' new requirements on the content, system, and activities of generative AI services will potentially impose considerable operational demands on companies developing generative AI systems and providing generative AI services to customers in China.
Further, Canadian businesses and organizations should take note of the speed and number of laws and regulations being implemented by Beijing. The Chinese government is actively developing a comprehensive regulatory regime for AI governance and the bureaucratic capacity to support the enforcement of these new regulations.
The Interim GAI Measures leaves regulators significant discretion, and the measures will require further clarity in their interpretation and application. Entities that could be subject to the Interim GAI Measures should consult with People's Republic of China-qualified legal counsel to ensure overall compliance and, where applicable, audit their existing policies and practices for compliance gaps. For general assistance with developing polices and governance frameworks for AI services, please contact our Privacy and Cybersecurity group or Technology, Media and Telecommunications group.
The authors would like to express their sincere appreciation to Justin P'ng, a former colleague at Fasken and Dongwoo Kim, summer student at Fasken, for their contributions to this article.
1. An unofficial translation of the Interim GAI Measures is available through China Law Translate: https://www.chinalawtranslate.com/en/generative-ai-interim/.
2. An unofficial translation of the draft measures (for solicitation of comments) is available through the DigiChina Project at Stanford University: https://digichina.stanford.edu/work/translation-measures-for-the-management-of-generative-artificial-intelligence-services-draft-for-comment-april-2023/.
3. "G7 leaders call for 'guardrails' on development of artificial intelligence", Financial Times, May 19, 2023: https://www.ft.com/content/1b9d1e21-ebc1-494d-9cce-97e0afd30c2d?desktop=true&segmentId=d8d3e364-5197-20eb-17cf-2437841d178a.
4. Citigroup, "China AI: What Lies Ahead?" Citigroup Inc., April 4, 2023: https://icg.citi.com/icghome/what-we-think/global-insights/insights/china-ai-what-lies-ahead.
5. An official translation is available through China's National Development and Reform Commission: https://en.ndrc.gov.cn/policies/202105/P020210527785800103339.pdf
6. An official translation is available through the government website of Fujian province: https://www.fujian.gov.cn/english/news/202108/t20210809_5665713.htm
7. An unofficial translation is available through the DigiChina Project at Stanford University: https://digichina.stanford.edu/work/full-translation-chinas-new-generation-artificial-intelligence-development-plan-2017/.
8. An unofficial translation is available through the DigiChina Project at Stanford University: https://digichina.stanford.edu/work/translation-internet-information-service-algorithmic-recommendation-management-provisions-effective-march-1-2022/.
9. An unofficial translation is available through China Law Translate: https://www.chinalawtranslate.com/en/deep-synthesis/#:~:text=Article%204%3A%20The%20provision%20of,i%E2%80%A6>
10. For more information on the Personal Information Protection Law, please see our bulletin on this topic: https://www.fasken.com/en/knowledge/2022/01/chinas-new-privacy-law-the-impact-on-canada-and-beyond.
11. This term is taken from the unofficial translation but is not otherwise defined.
The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.