China has made adjustments to its proposed regulations on generative AI, marking a shift in its approach to governing the emerging technology. The Cyberspace Administration of China, the country’s top internet watchdog, has recently released an updated version of its artificial intelligence guidelines. Notably, certain specific restrictions that were present in the initial draft, which was published in April and subjected to public comment, have been either removed or clarified.
The new regulations focusing on generative AI are expected to come into effect on August 15th. This move positions China as one of the first countries to establish regulatory frameworks for this nascent industry. Several leading Chinese internet companies such as Alibaba, Baidu, and JD have already announced their own AI bots as competitors to OpenAI’s ChatGPT
The original text has been translated from chinese
Title: Provisional Measures for the Management of Generative Artificial Intelligence Services
Date: July 20, 2023 (Thursday)
The State Internet Information Office, the National Development and Reform Commission, the Ministry of Education, the Ministry of Science and Technology, the Ministry of Industry and Information Technology, the Ministry of Public Security, and the State Administration of Radio and Television have jointly issued Order No. 15 - Provisional Measures for the Management of Generative Artificial Intelligence Services on May 23, 2023. The measures were reviewed during the 12th meeting of the State Internet Information Office in 2023 and approved by the National Development and Reform Commission, the Ministry of Education, the Ministry of Science and Technology, the Ministry of Industry and Information Technology, the Ministry of Public Security, and the State Administration of Radio and Television. The measures will take effect from August 15, 2023.
These measures aim to promote healthy development and regulated use of generative artificial intelligence technology, safeguard national security and public interests, protect the legitimate rights and interests of citizens, legal persons, and other organizations. The measures are based on relevant laws such as China’s Cybersecurity Law, Data Security Law, Personal Information Protection Law, Science and Technology Progress Law.
The measures apply to services that provide generative artificial intelligence content such as text, images audio or video to users within China. If there are specific regulations for activities related to news publishing, film production or literary creation using generative artificial intelligence services provided by industry organizations or research institutions that do not offer services to users within China.
The state adheres to a principle that emphasizes both development and security while promoting innovation guided by law. Effective measures are taken to encourage innovative development in generative artificial intelligence while implementing inclusive supervision based on classification management.
Providers and users must comply with applicable laws and regulations when providing or using generative artificial intelligence services. Providers must adhere to socialist core values and avoid generating content that incites subversion of state power, jeopardizes national security or interests, promotes division or social instability, spreads terrorism or extremism, incites ethnic hatred or discrimination, contains violent or obscene content, or disseminates false harmful information prohibited by laws and regulations. Furthermore, providers must take effective measures during the process of algorithm design, training data selection, model generation and optimization to prevent discrimination based on nationality, beliefs, region, gender, age, occupation or health status. Providers are also required to respect intellectual property rights and commercial ethics while protecting trade secrets. They must not engage in monopolistic practices or unfair competition using their algorithmic advantage.
Providers should respect the legitimate rights and interests of others and not harm their physical and mental health or infringe upon their portrait rights, reputation rights, honor rights, privacy rights and personal information rights. Providers should improve transparency in generative artificial intelligence services based on service types to enhance the accuracy and reliability of generated content.
The measures encourage innovation in the application of generative artificial intelligence technology across various industries while generating high-quality content that is positive and beneficial. Collaboration among industry organizations, companies, educational institutions, research institutions and public cultural institutions is supported in areas such as generative artificial intelligence technology innovation, data resource development and transformation application.
The measures also encourage independent innovation in generative artificial intelligence algorithms, frameworks, chips and supporting software platforms. International exchange and cooperation are promoted on an equal basis. The construction of generative artificial intelligence infrastructure and public training data resource platforms is encouraged. Collaborative sharing of computing power resources is promoted to improve utilization efficiency. Public data classification is expanded for orderly opening to ensure high-quality public training data resources. The use of secure and trusted chips, software tools, computing power resources is encouraged.
Providers who offer generative artificial intelligence services should follow certain rules when handling training data activities such as pre-training and optimization training. They should use data and basic models from legal sources, respect intellectual property rights, obtain personal consent or meet other legal requirements for personal information, take effective measures to improve the quality of training data, enhance its authenticity, accuracy, objectivity and diversity. Providers must also comply with relevant regulations and regulatory requirements from laws such as China’s Cybersecurity Law, Data Security Law, Personal Information Protection Law.
Providers who engage in data annotation during generative artificial intelligence technology development should establish clear, specific and actionable annotation rules that meet the requirements of these measures. They should conduct quality evaluations for data annotation, verify the accuracy of annotated content through sampling checks, provide necessary training for annotators to improve their awareness of abiding by laws and regulations, supervise and guide their work.
Providers are responsible for network information content production by relevant laws and regulations regarding the protection of personal information. Users’ input information and usage records must be protected in compliance with the law. Providers must not collect unnecessary personal information or illegally retain identifiable input information and usage records. They are forbidden from illegally providing others with users’ input information and usage records. Providers should promptly handle requests related to browsing, copying, correcting, supplementing or deleting personal information made by individuals in accordance with the law.
Providers must label images, videos and other generated content according to the Deep Synthesis Management Regulations for Internet Information Services.
Providers are obligated to provide secure, stable and continuous services during their provision of generative artificial intelligence services to ensure normal user experience.
If providers discover illegal content during service provision, they must take prompt measures such as stopping generation, transmission or removal as well as model optimization training to rectify the situation. If providers find that users are engaging in illegal activities using generative artificial intelligence services, they should take appropriate measures such as warnings or restrictions on functions or temporarily suspending or terminating services according to the law and contractual terms. Relevant records should be retained and reported to the relevant authorities.
Providers should establish a sound complaint and reporting mechanism, provide convenient channels for complaints and reports, publish handling procedures and feedback deadlines, timely accept and handle public complaints and reports, and provide feedback on the results of handling.
The management of generative artificial intelligence services will be strengthened by relevant departments such as the State Internet Information Office, the National Development and Reform Commission, the Ministry of Education, the Ministry of Science and Technology, the Ministry of Industry and Information Technology, the Ministry of Public Security, the State Administration of Radio and Television, and news publishing authorities in accordance with their respective responsibilities. The relevant competent authorities will improve regulatory methods suitable for innovative development based on scientific supervision in line with the characteristics of generative artificial intelligence technology applications in related industries.
If generative artificial intelligence services have features related to public opinion or social mobilization capabilities, they must undergo security assessments in accordance with relevant regulations. Providers must fulfill algorithm filing requirements according to “Regulations on Management of Internet Information Services Algorithm Recommendations” as well as notify changes or cancellations.
Users who find that generative artificial intelligence services do not comply with laws, regulations or these measures have the right to lodge complaints or reports with relevant authorities.
Relevant competent authorities are responsible for conducting supervision and inspections on generative artificial intelligence services based on their responsibilities. Providers must cooperate according to law, and explain training data sources, scale, types, annotation rules, and algorithm mechanisms as required by authorities during inspections. They shall also provide necessary technical support and assistance. Institutions and personnel involved in generative artificial intelligence service security assessments are obligated to keep state secrets confidentially as well as business secrets, personal privacy and personal information obtained through performing their duties. It is prohibited to disclose or illegally provide such information to others.
If generative artificial intelligence services provided from overseas to China do not comply with laws, regulations or these measures; the State Internet Information Office should notify relevant institutions to take technical and other necessary measures for disposal.
Violations of these measures by providers will be punished according to relevant laws and regulations such as China’s Cybersecurity Law, Data Security Law, Personal Information Protection Law, Science and Technology Progress Law. If there are no specific provisions in laws or regulations, the relevant competent authorities may issue warnings, criticism notices or order rectification based on their responsibilities. If providers refuse to make corrections or if the circumstances are serious, they may be ordered to temporarily suspend related services. Violations that constitute public security management offenses will be punished in accordance with public security management penalties. Violations that constitute crimes will be prosecuted according to law.
Definitions:
Generative artificial intelligence technology refers to models and related technologies capable of generating content such as text, images, audio or video.
Generative artificial intelligence service providers refer to organizations or individuals that provide generative artificial intelligence services (including through programmable interfaces).
Generative artificial intelligence service users refer to organizations or individuals that use generative artificial intelligence services to generate content.
These measures take effect from August 15, 2023.