ChatGPT has become one of the hottest topics around the world. Such an AI system is referred to as ‘Generative AI’, which could generate texts, images, voices, media, codes and other materials in response to what the user has imported or required, based on algorithms, models and rules. Generative AI may be used in various scenarios and promote the revolution of some industries, and thus it has aroused much attention. In response to the development of Generative AI, the Cyberspace Administrative of China (hereinafter “CAC”) issued the proposed Administrative Measures for Generative AI Service (Draft for Public Comment) (“生成式人工智能服务管理办法（征求意见稿）”, hereinafter “Draft Administrative Measures”) on 11 Apr. 2023.
The Draft Administrative Measures comprises 21 Articles, covering the key matters below.
1) Scope of Application: Extraterritorial Jurisdiction
This Draft Administrative Measures applies to anyone who innovates or utilizes Generative AI and provides services in China (the Providers) (Art 2). It means the Draft Administrative Measures would have an extraterritorial effect. Such an extraterritorial effect is based on the effect doctrine and the objective territorial doctrine. On the one hand, the content generated by Generative AI may affect the social order or other relationships in China, therefore having an effect in China. On the other hand, the Generative AI will be used by users in the territory, the operation of Generative AI located in the territory, which provides the basis of objective territorial doctrine.
2) Obligations of Providers
The Providers would take the following responsibilities pursuant to the Draft Administrative Measures.
The Providers should be responsible for the content generated by the AI. The content should be legal, accurate, and authentic. It should not be tortious, discriminative, fake, anti-social, anti-government, anti-national, terroristic, extreme, violent, pornographic and distractive of economic and social order. The Providers should delete the inappropriate content actively or at the request of users and improve or optimise the AI.
First, the Providers should comply with business ethics and should not carry out unfair competition conduct. Second, the Providers should not invade others’ rights, including intellectual property, personal information, privacy and trade secrets. Third, the Providers should not generate the content discriminatively.
The Draft Administrative Measures also provides requirements for the innovation of Generative AI. The providers should enact clear, specific and operable instruction for Contributors, train the Contributors and ensure the correctness of the content once the artificial annotation is required.
(4) Cooperation with the Administration
The Providers should actively cooperate with the administration required by the regulations or administrative agents. First, the Providers should register the Generative AI or report for security assessment according to related regulations on Internet service. Second, the Providers should mark the generated images and videos according to the Regulation on Service of Deepfake. Third, the Providers should provide essential information which could affect the choice of users at the request of the administrative agent.
(5) Management of the Users
The Providers should manage the users of Generative AI. First, the Providers should request the users to provide their identity information. Second, the Providers should specify and publicise the target of their service and take measures to prevent users from being addicted to the content provided by the AI. Third, the Providers should instruct the users to utilize the generative content appropriately and ban the users temporarily or permanently if they use the AI illegally or immorally.
1) Positive Effects
The Draft Administrative Measures is a quick response from China to the trend of Generative AI. It served as an essential supplement to the regulative framework of AI. It has several positive effects. First, it covers most of the legal issues brought by Generative AI, including the attribution of liability of AI, the threat to intellectual property, harmful content, unfair competition and the infringement of privacy, etc. Also, it provides whole-life supervision of the Generative AI, from innovation to utilization. Second, Article 3 of the Draft Administrative Measures has shown China’s supportive attitude towards the development of AI by declaring that the government would support and encourage the innovation, application and international collaboration of AI. Third, the Draft Administrative Measures serves as an essential part of the typological regulatory framework of AI, which consists of several regulations focused on different categories of AI.
2) Legal Status
The Draft Administrative Measures is not a law, since it is issued by CAC, an
Administrative Office under the State Council. According to the Legislation Law of China, laws can only be enacted by the National People’s Congress. Therefore, the CAC can only issue administrative rules or normative documents, which are at a lower hierarchy in China’s legislation hierarchy system and cannot conflict with laws. However, such a legal status has ensured flexibility and efficiency to some extent. On the one hand, compared with laws, the requirements to enact or amend the administrative rules or normative documents are less strict. On the other hand, All the administrative agents could issue regulatory documents within their own administrative authority when they consider it necessary, which ensures a quick response to the need for regulation.
It should be noted that there may be some other risks brought by the Draft Administrative Measures. Although Article 3 declared the supportive attitude of the government, it has to be acknowledged that the Administrative has significantly increased the burden on Providers of Generative AI. The technology of Generative AI is far from mature. Thus, it is doubtful whether the Providers are competent to comply with all the obligations provided by the Draft Administrative Measures, or at least it would increase the cost of compliance. For instance, the Draft Administrative Measures urges Providers to prevent fake content. However, technically speaking, the content can hardly be 100% authentic in the present stage. It may cost a lot for the Providers to comply with such an obligation. Accordingly, the high compliance burden may be an obstacle to the development and innovation of Generative AI to some extent.
To conclude, the Draft Administrative Measures reflects China’s positive attitude toward regulating AI. It has put a relatively high compliance requirement on entities intending to provide services of Generative AI. This attitude contributes to the healthy development of Generative AI, although there may be a concern about the high cost of compliance. Overall, the effect of the Draft Administrative Measures is still worthy of being observed.
 See http://www.cac.gov.cn/2023-04/11/c_1682854275475410.htm.
 Cyberspace Administrative of China, Ministry of Industry and Information Technology of China and Ministry of Public Security of China, Regulation on Service of Deepfake, Regulation No. 12.
Cover photo by Nebular Group on Unsplash
Contributors: Jidong Lin 林济东