# China AI Safety Regulations

China's Interim Measures for the Management of Generative AI Services, effective 15 August 2023, are regulations issued by the Cyberspace Administration of China and six other agencies to govern public-facing generative AI, balancing innovation with security and ethical requirements.

The Interim Measures for the Management of Generative AI Services (Chinese: 生成式人工智能服务管理暂行办法; pinyin: Shēngchéng shì réngōng zhìnéng fúwù guǎnlǐ zànxíng bànfǎ) are a set of regulations introduced by China to govern public-facing generative artificial intelligence within the country. These measures took effect on 15 August 2023, establishing a legal framework for the development, deployment, and supervision of generative AI systems that interact with the Chinese public. The regulations were issued jointly by the Cyberspace Administration of China and six other national regulators: 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 National Radio and Television Administration.

The measures represent one of the first comprehensive national regulatory attempts specifically targeting generative AI technologies, including large language models and other forms of [generative AI](https://www.wikiprompt.org/wiki/generative-ai). They aim to balance the promotion of technological innovation with the protection of national security, societal stability, and individual rights. The framework applies to services provided to the domestic public, while exempting internal corporate and research uses that do not reach public audiences.

## Scope and General Principles

The first section of the measures establishes the foundational scope and guiding principles. Article 1 states the intention to promote the healthy development and regulated usage of generative AI while safeguarding national security, societal public interests, and the legal rights of citizens and organizations. Article 2 defines the scope as applying to public-facing generative AI services within the People's Republic of China, explicitly exempting internal services by companies, research institutions, and other organizations that do not provide services to the domestic public.

Article 3 sets out key regulatory principles: equal emphasis on development and security, promoting innovation alongside the rule of law, and adopting an inclusive and prudent approach with classified supervision of generative AI services. This dual focus on development and security reflects a broader policy approach in China's technology governance.

Article 4 imposes substantive content requirements. Generative AI services must adhere to relevant laws and respect social morality, aligning with the Core Socialist Values. The measures prohibit content that promotes overthrowing the socialist system, harms national security, or undermines social stability. They also ban content that promotes discrimination, terrorism, extremism, violence, obscenity, or false and harmful information prohibited by law. Models must be designed and trained to avoid discrimination or algorithmic bias, and services must respect intellectual property rights and business ethics. Providers are required to protect others' rights, avoid endangering physical and mental health, and take effective measures to improve transparency and accuracy of their services.

## Technology Development and Governance

The second section focuses on encouraging innovation while establishing governance requirements for the underlying technology. Article 5 encourages innovative application of generative AI across industries to create positive, healthy, and high-quality content, supporting collaboration on AI innovation, data, and application development. Article 6 promotes independent innovation in core AI technologies, infrastructure development, and data resource platforms, while encouraging international cooperation, collaborative sharing of compute resources, expansion of high-quality public training data sources, and adoption of secure chips and resources.

Article 7 mandates that providers employ effective measures to ensure training data is high-quality, accurate, diverse, and legitimately sourced without infringing intellectual property rights. When personal information is used in training, providers must obtain user consent. This provision directly addresses concerns about data provenance and privacy in the development of [machine learning](https://www.wikiprompt.org/wiki/machine-learning) systems.

Article 8 requires providers to formulate clear, specific, and operable data labeling rules in line with the measures. It mandates quality assessment, sampling verification, personnel training, and supervision for labeling work, recognizing the importance of data quality in training reliable AI models.

## Service Specifications

The third section details obligations for providers when offering services to the public. Article 9 legally establishes providers as producers of online content and processors of personal data, requiring them to sign service agreements with users that define rights and obligations. Article 10 mandates taking effective measures to prevent excessive use by minors, requiring providers to clarify appropriate use and help users understand generative AI.

Article 11 obligates providers to protect user privacy, minimize data collection and retention of personal information, and avoid illegally giving user information to others. Providers must accept user requests to review, copy, modify, or delete their personal information, aligning with broader data protection principles. Article 12 requires labeling of synthesized pictures, videos, and other generated content, addressing concerns about deepfakes and misinformation.

Article 13 requires stable and continuous services for normal usage by users. Article 14 obligates providers to promptly address situations where a service generates illegal content or is used for illegal activities, potentially including halting generation, retraining a model, warning users, and reporting to relevant authorities. Article 15 mandates complaint and reporting mechanisms for public grievances with timely processing and feedback.

## Supervision and Legal Liability

The fourth section establishes the supervisory framework and legal consequences. Article 16 gives regulatory oversight roles to cybersecurity, technology, industry, education, public security, and other relevant government bodies. Article 17 requires security assessments for services with public opinion influence, along with algorithm filing procedures, creating a dual mechanism of pre-deployment review and ongoing registration.

Article 18 allows users to complain to authorities about non-compliant services. Article 19 enables inspections of providers by officials, requiring cooperation and transparency. Article 20 allows the government to address foreign services that violate the measures, extending the regulatory reach to international providers operating in China. Article 21 defines the relevant authorities that may punish providers that violate the provisions, with crimes investigated according to law.

## Supplementary Provisions

The fifth section contains definitions and implementation details. Article 22 defines key terms such as generative AI, service providers, and service users. Article 23 requires obtaining relevant licenses to provide services where required by law. Article 24 establishes that the measures take effect on 15 August 2023.

The measures apply to a wide range of generative AI technologies, including those built on [large language models](https://www.wikiprompt.org/wiki/large-language-model) and [transformer architectures](https://www.wikiprompt.org/wiki/transformer). They affect both domestic Chinese companies and foreign firms offering services in China, requiring compliance with content standards, data governance, and security assessment procedures.

## Implementation and Impact

The implementation of these measures has significant implications for the AI industry. Providers must conduct security assessments before launching services that could influence public opinion, and they must file their algorithms with authorities. The requirement for data labeling rules and quality assessment affects how training datasets are constructed and validated.

The measures also address the use of [deep learning](https://www.wikiprompt.org/wiki/deep-learning) techniques and [neural networks](https://www.wikiprompt.org/wiki/neural-network) in generative systems. While the regulations do not mandate specific technical approaches, they require that models avoid algorithmic bias and discrimination, which influences model design and training practices.

For international AI companies, the measures create compliance obligations when offering services to the Chinese public. This includes adapting content moderation systems, implementing user consent mechanisms for personal data use, and cooperating with government inspections. The provisions on foreign services indicate that extraterritorial enforcement is possible for violations.

The regulatory approach reflects a broader trend in China toward comprehensive AI governance. The measures complement other regulations on cybersecurity, data protection, and algorithm recommendation systems, creating a layered regulatory environment. The emphasis on both development and security suggests an ongoing effort to maintain technological competitiveness while managing potential risks.

As of the effective date, the measures represent a baseline framework that may be updated as generative AI technologies evolve. The requirement for classified supervision suggests that different types of services may face different levels of scrutiny depending on their potential impact on public opinion and social stability. The measures also encourage international cooperation in AI innovation, indicating an openness to global collaboration despite the regulatory constraints.

The practical implementation of these measures continues to develop, with authorities issuing guidance and conducting assessments. The impact on AI research and development in China includes increased attention to data quality, labeling practices, and content safety measures. The measures have also influenced discussions about AI governance in other jurisdictions, contributing to the global conversation on how to regulate generative AI technologies.

## See Also

- [Artificial intelligence](https://www.wikiprompt.org/wiki/artificial-intelligence)
- [Generative AI](https://www.wikiprompt.org/wiki/generative-ai)
- [Large language model](https://www.wikiprompt.org/wiki/large-language-model)
- [Machine learning](https://www.wikiprompt.org/wiki/machine-learning)

---
Source: https://www.wikiprompt.org/wiki/china-ai-safety-regulations
License: CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/)
Last updated: 2026-09-12T16:23:44.279517+00:00