The People’s Republic of China (“China”) is positioned at the forefront of AI-driven technological developments. China’s AI market, as of 2021, is worth about 150 billion RMB (23.196 billion USD).¹ China is projected to receive a 26% boost to GDP in 2030 from AI-powered economic gains and contribute 30 trillion RMB’s (4.25 trillion USD) worth of economic value by 2035.²

This article summarizes China’s recent AI regulatory trends and provides considerations for businesses exploring AI-related opportunities in China.

Overview

AI technology is booming in China. Since 2016, China has led in global AI patent filings, accounting for 61.1% of AI patent origins in 2022.³  The rapid growth of AI research and deployment in China has attracted funding from global private equity and venture capital—one-fifth of such investments went to AI start-ups in China in 2021.⁴

The growth of AI technology has brought great benefits but raised concerns at the same time. For example, AI service providers have been criticized for engaging in differential pricing, which are generated based on users’ demographic information and purchasing behavior.⁵

To address AI-related concerns, the Chinese government has implemented non-mandatory guidelines as well as mandatory regulations that target leading AI technologies, including but not limited to, recommendation algorithms, deepfakes, and generative AI.

AI Governance in Mainland China: A Timeline

I. Non-mandatory Guidelines

AI governance in China began with the release of the “New Generation Artificial Intelligence Development Plan” (新一代人工智能发展规划) in 2017. The 2017 Plan outlines AI development goals for various sectors and proposes an AI ecosystem consisting of research and development, technology applications, and industrial upgrading. It also puts forth a regulatory framework that focuses on traceability, accountability, and international collaboration.

Between 2017 and 2020, AI regulation in China emphasized self-governance in companies and organizations. For instance, Several AI companies based in Shenzhen, the epicenter of China’s AI and tech development, launched the “New-Generation AI Industry Self-Regulation Convention” (新一代人工智能行业自律公约) in 2019, following the issuance of “AI Industry Self-Regulation Convention” (人工智能行业自律公约), an initiative by the Artificial Intelligence Industry Alliance (AIIA), a major coalition of Chinese AI companies.⁶

In 2020, the Cybersecurity Administration of China (CAC) and the Ministry of Industry and Information Technology jointly issued the “Guidelines for the Construction of a National New-Generation AI Standard System” (国家新一代人工智能标准体系建设指南). The 2020 Guidelines provides sector-specific standardization guidance for AI deployment. It also identifies risks associated with AI applications and prescribes codes of practice for the fair use of data, algorithms, and modeling.

In 2021, several AI-related guidelines were published. In September of that year, the Ministry of Science and Technology issued “Ethical Norms for New Generation Artificial Artificial Intelligence” (新一代人工智能伦理规范) to address ethical risks associated with AI deployment. The Personal Information Protection Law of the People’s Republic of China (中华人民共和国个人信息保护法) and Data Security Law of the People’s Republic of China (中华人民共和国数据安全法) were also enacted in 2021 in response to data breaches on AI-powered online platforms.

Since 2021, the Chinese government has become more active in adopting mandatory regulations. The sections below analyze three key AI mandatory regulations.

II. Mandatory Regulation

2021 Regulations on Recommendation Algorithms

In response to the rise of algorithm-driven news aggregators, the Cybersecurity Administration of China (CAC) released the Provisions on the Management of Algorithmic Recommendations in Information Services (互联网信息服务算法推荐管理规定, “Algorithmic Recommendation Provisions”) in 2021. TouTiao, the largest news aggregator in China, surged in popularity in 2016 because of its powerful recommendation algorithms that tailor content based on users’ interests.⁷ But TouTiao’s popularity raised data privacy and consumer protection concerns:  To attract users, TouTiao collected a great amount of user data and curated articles based on a wide range of sources, many of which supplied fake news catering to users’ interests, or click-bait leading to scams.⁸

The Algorithmic Recommendation Provisions addressed these concerns by regulating the building blocks of recommendation algorithms involving data generation and synthesis, selection sorting, search filtering, and individualized “push”.⁹ Under the Provisions, the recommendation algorithms providers are not allowed to block information, manipulate search results, or make excessive recommendations.¹⁰ Additionally, users have rights to (1) turn off algorithmic recommendations, (2) select or delete user tags for personalizing content recommendations and (3) obtain explanations of algorithms from service providers.¹¹

A novel creation of the 2021 Provisions is the Internet Information Service Algorithm Filing System (互联网信息服务算法备案系统). The Algorithm Filing System is an algorithm governance registry created by the CAC to track algorithms that “shape public opinion” or “mobilize society.”¹² A wide variety of internet platforms—ranging from financial products to social media apps—have filed with the System between 2022 and 2024.¹³

To file with the System, algorithm providers are required to disclose whether their data inputs and outputs involve biometric or other personal information.¹⁴ Providers are encouraged, but not required, to provide details of algorithmic modeling, including each open-source and self-built data set used to train the model, as well as the source of the data.¹⁵ In addition, filers must submit an “Algorithm Security Self-Assessment Report” that includes measures implemented by the platforms to address potential risks related to the use of recommendation algorithms.¹⁶ Only a high-level description of the registered algorithms would be disclosed to the public.¹⁷

2022 Regulations on “Deep Synthesis”

In 2022, the Chinese regulators turned their attention to another serious issue in AI: deepfakes. Deepfakes are manipulated content depicting seemingly real people or events.¹⁸  Deep learning applications like ZAO, a face-swapping app where users could swap the faces of celebrities onto the bodies of other people caused widespread debates about data privacy and likeness rights.¹⁹ Tencent, one of China’s premier tech giants, issued a report in which it distinguishes deepfakes from deep learning mechanisms and advocated for using “deep synthesis” (深度合成) as a term of art for AI-generated content to dispel mistrust towards emerging technologies.²⁰

Soon after the release of Tencent’s report, the CAC issued the Internet Information Service Deep Synthesis Management Provisions (互联网信息服务深度合成管理规定). The 2022 Deep Synthesis Provisions defines deep synthesis technologies as “generative sequencing algorithms to create text, images, audio, video, virtual scenes, or other information.”²¹ The 2022 Provisions covers applications including chat bots, voice conversion, music generation, face swapping, and 3D reconstruction.²²

Under the 2022 Provisions, deep synthesis service providers are required to attach a “conspicuous label in a reasonable position” on AI-generated content to alert the public to the use of deep synthesis technologies.²³ The providers must obtain consent from the users “whose personal information is being edited.”²⁴ Additionally, the algorithms providers must verify the identity of users and decline access to users who fail to register with their real names.²⁵ In line with the 2021 Recommendation Algorithms Provisions, the service providers must register their algorithms with the Algorithm Filing System.²⁶

2023 Regulations on Generative AI

The release of ChatGPT in November 2022 shook the world with its powerful large language models (“LLMs”). In response to the rapid growth of generative algorithms, the CAC, along with six other Chinese agencies, issued Measures for the Management of Generative Artificial Intelligence Services (生成式人工智能服务管理办法, “Generative AI Measures”) in 2023 to regulate all forms of generative AI applications, including but not limited to the synthesized visual and audio content that had already been covered under the 2022 Provisions.

Building upon the 2022 Deep Synthesis Provisions, the 2023 Generative AI Measures strikes a more intricate balance between promoting innovation and controlling risks. This is reflected in several significant changes to the final 2023 Measures in comparison to the early draft.

First, the final version includes an explicit exemption for enterprises and institutions that research, develop, and use generative AI technology but do not provide generative AI services to the mainland public.²⁷ Specifically, the Generative AI Measures would not apply to the internal use of generative AI technologies, as well as generative AI services provided by certain entities to business operations outside of China.²⁸ This narrowed the scope in contrast with previous regulations, which covered any AI service providers in China.²⁹

Second, the final version removes an explicit restriction on using user input data and usage patterns for profiling purposes.³⁰ Instead, the providers’ liability is limited to collecting “unnecessary personal information.”³¹ The “minimum necessary” standard seems to grant AI service providers more leeway to track user behavior.

Finally, the final version removes the requirement that providers need to ensure the “authenticity, accuracy, objectivity, and diversity of training data” and requires only that the providers “employ effective measures to increase the quality of training data.”³² Additionally, the 2023 Measures removes penalties in the form of service termination or fines in an amount between 10,000 to 100,000 RMB (1,415 USD to 14,150 USD).³³

Takeaways

AI regulations in China closely track the domestic public discourse about technology risks— Mainland China has enhanced state-guided policymaking in AI as the public has become more anxious about AI risks. And, more legally binding requirements will likely appear in the future. Yet, regulators in China are sensitive to the economic potential of AI, and thus have attempted to refine regulatory measures to accommodate business growth. By using the Algorithm Filing System to better understand latest trends in AI development and deployment, regulators in China are showing responsiveness to feedback from the tech community and willingness to work with AI service providers to foster the AI economy in an informed manner.

About the Author

Sophia Wang is a Summer Associate in HM.

1 Yi Wu, AI in China: Regulations, Market Opportunities, Challenges for Investors, China Briefing. October 14, 2022.

2 Ben Jiang, Chinese generative AI to account for a third of industry’s economic value by 2035, Beijing think tank says, South China Post. December 15, 2023.

3 Standford University Human-Centered Artificial Intelligence, Artificial Intelligence Index Report 2024.

4 Kai Shen et al., The Next frontier for AI in China could add $600 billion to its economy, QuantumBlack AI by McKinsey. June 7, 2022.

5 Ye Zhao, Price Discrimination on Chinese E-Commerce Platforms – Abuse of Algorithmic Power, Digital Policy & Governance Blog. April 14, 2024.

6 Ngor Luong and Zachary Arnold, China’s Artificial Intelligence Industry Alliance: Understanding China’s AI Strategy Through Industry Alliances, Center for Security and Emerging Technology (May 2021), 

7 Matt Sheehan, Tracing the Roots of China’s AI Regulations, Carnegie Endowment for International Peace. Feb. 27, 2024.

8 Elliott Zaagman, Updated: Toutiao’s overseas platform is delivering fake news, but its problems run much deeper than that. Feb. 1, 2018.

9 Rogier Creemers et al., Translation: Internet Information Service Algorithmic Recommendation Management Provisions. Jan. 10, 2022.

10 China Law Translate, Provisions on the Management of Algorithmic Recommendations in Internet Information Services. Aug. 27, 2021.

11 Ibid.

12 Ibid.

13 Algorithm Filing System (互联网信息服务算法备案系统).

14 Ibid.

15 Matt Sheehan, What China’s Algorithm Registry Reveals about AI Governance. Dec. 9, 2022. 

16 Ibid.

17 Ibid.

18 Meredith Somers, Deepfakes, explained. July 21, 2020.

19 Agence France-Presse, Chinese deepfake app Zao sparks privacy row after going viral. Sept. 2, 2019.

20 On ten misconceptions about “deep synthesis” technology (关于”深度合成”技术的十个误解).

21 China Law Translate, Provisions on the Administration of Deep Synthesis Internet Information Services. Nov. 25, 2022.

22 Ibid.

23 Ibid.

24 Ibid.

25 Ibid.

26 Ibid.

27 Jeremy Daum, Key Changes to Generative AI Measures. July 18, 2023.

28 China Law Translate, Comparison Chart of Current vs. Draft rules for Generative AI. July 13, 2023.

29 Yirong Sun & Jingxian Zeng, China’s Interim Measures For The Management Of Generative AI Services: A Comparison Between The Final and Draft Versions of The Text. Apr. 22, 2024.

30 Jeremy Daum, Key Changes to Generative AI Measures. July 18, 2023.

31 China Law Translate, Comparison Chart of Current vs. Draft rules for Generative AI. July 13, 2024. 

32 Ibid.

33 Ibid.