The Latest Draft Measures on the Management of Generative AI

Tag：intellectual property, digital economy, telecommunications media entertainment and technology-technology,artificial-intelligence

In response to the rapid development of Generative AI services in the past few months, especially the launch of ChatGPT since last December, on April 11, 2023, Chinese Cyberspace Administration issued a draft policy Measures on the Management of Generative Artificial Intelligence Services (“Draft AI Policy”), soliciting feedback from the general public with respect to the regulation and management of Generative AI services.

The Draft AI policy addressed a few important topics surrounding AI and AIGC, including (1) the definition of Generative AI and to what extend this Draft AI Policy should apply; (2) whether text and data mining (“TDM”) might constitute copyright infringement in training AI; (3) misinformation risks associated with AIGC; (4) data privacy issue; and last but not least (5) security evaluation of AI service provider and its filing procedure etc.

1. Generative AI and the Scope of this Draft AI Policy

To start with, "Generative AI" is defined in the Draft AI Policy as a technology that generates text, images, sound, video, codes and other contents based on algorithms, models, and rules.

Article 2 of the Draft AI Policy stipulates that, regardless whether a Generative AI service or product is developed or used in China, as long as it is targeting at consumers located within China, that AI service or product provider will be subject to the Draft AI Policy.

2. TDM Issue Training AI

Unlike the UK, EU, Japan or Singapore that allows text and data mining exception in training AI, the Draft AI Policy explicitly states in Article 7 that, the AI product/service provider shall be responsible for the “legality of the source of pre-training and optimization training data”, thus requiring the AI product/service provider to obtain clearance from IP rights holders prior to using such copyrighted works to train AI.  Given the limited few statutory categories of “fair use” exceptions in Chinese Copyright Law and the mixed results of copyright fair use cases in the recent Chinese case law, the use of copyrighted works to train AI (i.e. TDM) still faces big challenges of copyright infringement suits.

3. Misinformation

Chinese legislatives are apparently well aware of the risks associated with AIGC, in terms of potential discrimination issue, misinformation and violation of right of publicity etc. Specifically, the Draft AI Policy (Articles 4, 7 and 12) requires AI product/service providers to apply “technical means” to “avoid the generation of illegal content, false information, and discriminatory content as much as possible.”

4. Data Privacy

In response to the data privacy concerns associated with generative AI, the Draft AI Policy addresses the issue. Specifically, it emphasizes the obligation of AI product/service providers to protect users' personal information and privacy. Article 11 of the Draft AI Policy sets a limit on information retention for generative AI products/services, stipulating that "providers shall not illegally retain input information that can infer the user’s identity, shall not draw portraits based on the user’s input information and usage, and shall not provide the user’s input information to others." Article 13 of the Draft AI Policy also requires AI product/service providers to establish a mechanism for receiving and handling user complaints, stating that "measures shall be taken to stop such generation and prevent enduring hazard" when an infringement is discovered. 

5. Security Evaluation and Filing Procedure

Article 6 of the Draft AI Policy requires that any release of AIGC products must undergo a security evaluation and be filed with the competent authorities. Article 6 stipulates that “Before providing services to the public with generative artificial intelligence products, it shall report to the National Cyberspace Administration for security assessment in accordance with the Regulations on the Security Assessment of Internet Information Services with the Attributes of Public Opinion or the Ability of Social Mobilization, and go through the procedures of algorithm filing, alteration and cancellation in accordance with the Regulations on the Management of Algorithmic Recommendations in Internet Information Services.”

6. Conclusion

It should also be noted that the Draft AI Policy is silent on the copyrightability of AIGC, i.e., whether AI generated content should be entitled to copyright protection, which is traditionally reserved for human authors. It is unclear whether Chinese Copyright Office will issue a separate policy statement on the copyrightability of AIGC, if it is not to be covered under this Draft AI Policy. In view of the recent two cases[1] on AIGC, we feel that Chinese attribute on the copyrightability of AIGC might be similar to the copyright policy statement[2] issued by the U.S. Copyright Office on March 16, 2023. This Draft AI Policy is one of the recent legislative efforts that Chinese government made to regulate the development of AI technology and services. In the Regulations on the Management of Algorithmic Recommendations in Internet Information Services and the Regulations on the Administration of Deep Synthesis Internet Information Services issued in 2022, Chinese government also addressed the issues of deep fake and data privacy associated with AI development.

(Attached is the full English translation of the Draft AI Policy prepared by CHAN PuiLing at KWM)

The Measures for the Administration of Generative Artificial Intelligence Service

(Draft for comments)

Article 1 For the purpose of promoting the sound development and standardized application of generative artificial intelligence, the Measures are formulated in accordance with the Cybersecurity Law of the People's Republic of China, the Data Security Law of the People's Republic of China, the Personal Information Protection Law of the People's Republic of China and other laws and administrative regulations.

Article 2 The Measures shall apply to those who develop and utilize generative artificial intelligence products to provide services to the public within the territory of the People's Republic of China.

Generative artificial intelligence as mentioned in the Measures refers to the technology of generating text, image, sound, video, code and other contents based on algorithms, models and rules.

Article 3 The State supports independent innovation, popularization and application of artificial intelligence algorithms, frameworks and other basic technologies as well as international cooperation, and encourages priority to be given to the use of secure and trustworthy software, tools, calculations and data resources.

Article 4 The provision of generative artificial intelligence products or services shall comply with the requirements of laws and regulations, observe social ethics, public order and good morals, and meet the following requirements:

(1) The content generated by generative artificial intelligence should demonstrate the socialist core values. It must not contain content such as subverting state power, overthrowing the socialist system, inciting secession of the country, undermining national unity, promoting terrorism and extremism, promoting ethnic hatred and discrimination, violence, obscene and pornography, false information, and content that may disturb economic and social order.

(2) In the process of algorithm design, training data selection, model generation and optimization, and service provision, measures should be taken to prevent discrimination based on race, ethnicity, belief, nationality, region, sex, age, occupation, etc.

(3) Respect intellectual property rights and business ethics, and shall not use algorithms, data, platforms and other advantages to carry out unfair competition.

(4) The content generated by generative artificial intelligence shall be authentic and accurate, measures should be taken to prevent the generation of false information.

(5) Respect the legitimate interests of others, prevent injury to the physical and mental health of others, prevent damage to the right of likeness, right of reputation and personal privacy, and the infringement of intellectual property rights. It is prohibited to illegally obtain, disclose or use personal information, privacy and business secrets.

Article 5 Organizations and individuals (hereinafter referred to as "providers") that use generative artificial intelligence products to provide services such as chat, text, image and sound generation, including supporting others to generate text, image and sound by providing programmable interfaces, shall assume the responsibilities of producers of the content generated by such products; Where personal information is involved, it shall bear the statutory responsibility of the person handling personal information and fulfill the obligation of personal information protection.

Article 6 Before providing services to the public with generative artificial intelligence products, it shall report to the National Cyberspace Administration for security assessment in accordance with the Regulations on the Security Assessment of Internet Information Services with the Attributes of Public Opinion or the Ability of Social Mobilization, and go through the filing, alteration and cancellation procedures for algorithm in accordance with the Regulations on the Management of Algorithmic Recommendations in Internet Information Services.

Article 7 Providers shall be responsible for the legality of pre-training and optimized training data sources of generative AI products.

The pre-training and optimized training data used for generative AI products shall meet the following requirements:

(1) It should comply with the requirements of the Cybersecurity Law of the People's Republic of China and other laws and regulations;

(2) The contents should not infringe intellectual property rights;

(3) If the data contains personal information, consent of the subject of personal information shall be obtained or other circumstances prescribed by laws and administrative regulations shall be complied with;

(4) The authenticity, accuracy, objectivity and diversity of data should be ensured;

(5) Other regulatory requirements of the Cyberspace Administration of China on generative artificial intelligence services.

Article 8 When manual labeling is used in the development of generative artificial intelligence products, providers shall make clear, specific and operable labeling rules in accordance with the requirements of the Measures, conduct necessary training for labeling personnel, and verify the correctness of labeling contents by sampling.

Article 9 The provision of generative artificial intelligence services shall require users to provide real identity information in accordance with the provisions of the Cybersecurity Law of the People's Republic of China.

Article 10 Providers shall clarify and disclose the applicable groups, circumstances, and uses of its services, and take appropriate measures to prevent users from excessively relying on or indulging in the generated content.

Article 11 In the process of providing services, providers shall undertake the obligation to protect the input information and usage records of users. Providers shall not illegally retain input information that can infer the identity of the user, shall not draw portraits based on the user’s input information and usage, and shall not provide the user’s input information to others, except as otherwise provided for by any law or regulation.

Article 12 Providers shall not generate discriminatory content based on the user's race, nationality, sex, etc.

Article 13 Providers shall establish a mechanism for receiving and handling users' complaints and promptly handle requests for correction, deletion or shielding of their personal information; When generated text, image, sound or video is found to infringe upon the right of likeness, right of reputation, personal privacy and trade secrets of others, or fail to meet the requirements of the Measures, measures shall be taken to stop such generation and prevent enduring hazard.

Article 14 Providers shall, within the life cycle, provide safe, robust and continuous services to ensure the normal use of users.

Article 15 For the generated content found in operation or reported by users that does not meet the requirements of the Measures, in addition to taking measures such as content filtering, it shall be prevented from being regenerated by means of model optimization training within 3 months.

Article 16 Providers shall mark the generated images, videos and other contents in accordance with the Provisions on the Administration of Deep Synthesis of Internet-based Information Services.

Article 17 Providers shall, in accordance with the requirements of the National Cyberspace Administration and relevant authorities, provide necessary information that can affect the trust and choice of users, including the description of the source, scale, type and quality of pre-training and optimized training data, manual labeling rules, scale and type of manual labeling data, basic algorithm and technical system, etc.

Article 18 Providers shall guide users to have a scientific understanding and rational use of the content generated by generative artificial intelligence, not to damage the image, reputation or other legitimate rights and interests of others by using the generated content, and not to engage in commercial speculation or improper marketing.

When users find that the generated content does not meet the requirements of the Measures, they shall have the right to report to the National Cyberspace Administration or relevant authorities.

Article 19 Providers shall suspend or terminate the service when it finds the user’s violation of laws and regulations, business ethics or social morality in the process of using the generative artificial intelligence products, including engaging in network speculation, malicious posting and commenting, creating spam, programming malicious software, and implementing improper commercial marketing, etc.

Article 20 Providers who violate the provisions of the Measures shall be punished by the National Cyberspace Administration and relevant authorities in accordance with the Cybersecurity Law of the People's Republic of China, the Data Security Law of the People's Republic of China, the Personal Information Protection Law of the People's Republic of China and other laws and administrative regulations.

Where laws or administrative regulations do not provide for it, the National Cyberspace Administration and relevant authorities shall, in line with their functions and duties, issue warnings, circulate notice of criticism and order correction within limited time; Where corrections are refused or the circumstances are serious, the offender shall be ordered to suspend or terminate the use of generative artificial intelligence to provide services and be fined not less than 10,000 yuan but not more than 100,000 yuan. If the case constitutes an act violating the administration of public security, penalties for the administration of public security shall be imposed in accordance with law; If the case constitutes a crime, criminal responsibility shall be pursued in accordance with law.

Article 21 The Measures shall be implemented as of [day][month] 2023.

Thanks to CHAN PuiLing and Wang Mo for their contributions to this article.

Scan the QR code to subscribe to "King & Wood Mallesons" for more information