China is actively working towards regulating generative AI while concurrently focusing on creating a supportive ecosystem for AI-driven innovation, scientific and technological advancements, and R&D applications in various sectors. The recently introduced draft Measures for Generative Artificial Intelligence Services aims to ensure that industry outcomes align with China’s fundamental socialist values, social ethics, and public order.
Essentially, the goal is to strike a balance between promoting innovation within the AI sector and upholding ethical considerations.
In April, the draft “Measures for Generative Artificial Intelligence Services” (《生成式人工智能服务管理办法（征求意见稿）》, referred to as the “draft Measures”) was released by the Cyberspace Administration of China. These measures are specifically designed to regulate generative artificial intelligence (AI) products such as ChatGPT and establish guidelines that govern the operation of generative AI services, including the permissible content that these products can generate.
The draft Measures also address key concerns of the Chinese government regarding the utilization of generative AI, such as content moderation, prevention of information distortion and abuse, mitigation of algorithmic bias and prejudice, and the promotion of transparency.
China’s regulations on generative AI
In recent years, China has positioned itself as a leader in AI regulations by implementing a range of initiatives aimed at nurturing and overseeing the expansion of the AI sector. Noteworthy examples include Made in China 2025, the Action Outline for Promoting the Development of Big Data (2015), and the Next Generation Artificial Intelligence Development Plan (2017).
China has demonstrated proactive measures in enacting laws that govern the ethical practices of AI businesses and algorithms. As part of their broader endeavors to regulate the technology industry, it is conceivable that the Chinese government may introduce regulations pertaining to AI-based language models, such as HunyuanAide or Baidu Ernie Bot.
Regarding the draft Measures, there is a possibility that the regulation will be passed and take effect later this year, complementing China’s existing regulatory frameworks. These frameworks encompass various laws and regulations such as the Cybersecurity Law, Data Security Law, Personal Information Protection Law (PIPL), Deep Synthesis Provisions, and Administrative Provisions on Recommendation Algorithms in Internet-based Information Services. Together, they provide guidance for the development of the AI industry.
In essence, the draft Measures aim to mitigate the risks and potential misuse of AI technologies while promoting AI innovation that aligns with China’s core socialist values, social ethics, and public order. By integrating these regulations, China strives to strike a balance between fostering advancements in AI and ensuring responsible and ethical AI practices within the country.
Furthermore, despite the requirements set forth in the draft Measures, the document emphasizes the government’s encouragement for the domestic development and international cooperation concerning generative AI technology. It also encourages companies to adopt “secure and trustworthy software, tools, computing, and data resources” to facilitate these objectives.
Applicability
The draft Measures define generative AI as technologies that generate text, pictures, sounds, videos, codes and other content based on algorithms, models, and rules. They specifically regulate the broader category of generative AI technologies, as opposed to merely “deep synthesis” technologies. If enacted, the draft Measures will regulate the activities of service providers offering their products to users in mainland China, regardless of whether the service is based locally or abroad.
Subscribe to our newsletter and receive regular information and insights.
“Provider of generative AI” and governance
In the draft Measures, a “provider of generative AI” is defined as organizations and individuals who utilize generative AI products to offer services encompassing chat, text, image, and sound generation. This definition includes those who facilitate content generation through interfaces and platforms. Notably, the draft Measures do not distinguish between back-end technology providers and application-level service providers. Both types of providers bear responsibility for the content generated by generative AI products and are obligated to adhere to China’s Personal Information Protection Law (PIPL) in order to safeguard personal information.
The draft Measures also aim to establish regulatory control over generative AI technologies by imposing specific obligations on the “Providers.” These Providers hold the responsibility of ensuring the authenticity of the data utilized for training generative AI products, as well as the generated content itself. They are obligated to submit security assessment reports to the relevant authority and adhere to procedures for algorithm filing, modification, and cancellation prior to offering services to the public involving the use of generative AI products. The draft Measures outline explicit requirements for service providers, emphasizing their accountability for the accuracy and truthfulness of the generated content.
Requirements
The draft Measures for Generative Artificial Intelligence Services reinforce existing legal obligations and introduce additional requirements for providers of generative AI products and services.
These requirements encompass:
- Implementing measures to prevent discriminatory practices based on factors such as race, ethnicity, religion, nationality, geography, sex, age, occupation, and more.
- Respecting the intellectual property rights of others and refraining from engaging in unfair competition by utilizing algorithms, data, or platforms.
- Taking steps to ensure the authenticity of generated information and preventing the dissemination of disinformation.
- Safeguarding input information and user history, prohibiting the unauthorized retention of user-identifying input data, abstaining from creating user profiles based on such data, and refraining from sharing user input information with third parties.
- Requiring users to provide genuine identities and relevant information, while refraining from generating false or deceptive information.
- Suspending or discontinuing services in instances where they are exploited for activities that contravene commercial or social ethics, such as engaging in misleading promotions, malicious postings or comments, spam creation, development of malicious software, or engaging in inappropriate commercial marketing practices.
- Clearly delineating the target user base, specific usage scenarios, and permissible applications of their services, while implementing measures to prevent addiction.
- Furnishing necessary information to the Cybersecurity Administration of China (CAC) upon request, which includes details such as dataset sources, size, characteristics, quality, labeling guidelines, specific labeled dataset information, underlying algorithms, and technical systems.
Users’ rights according to the draft Measures
Violations of the Measures will be subject to penalties in accordance with relevant laws. If there is no specific provision for the penalty, the CAC and other applicable regulators may issue warnings, require corrections within a specified timeframe, order the suspension or termination of services, and impose fines ranging from RMB 10,000 to 100,000 (US$1,470 to US$14,700). In cases where inappropriate content is generated, providers must update their technology within three months to prevent its recurrence. These measures aim to ensure compliance, protect personal information, and address content-related issues promptly.
Liability for non-compliance and breaches
Violations of the Measures will be subject to penalties in accordance with relevant laws. If there is no specific provision for the penalty, the CAC and other applicable regulators may issue warnings, require corrections within a specified timeframe, order the suspension or termination of services, and impose fines ranging from RMB 10,000 to 100,000 (US$1,470 to US$14,700). In cases where inappropriate content is generated, providers must update their technology within three months to prevent its recurrence. These measures aim to ensure compliance, protect personal information, and address content-related issues promptly.