This page summarizes the English translation of China's interim regulation governing public-facing generative AI services, promulgated 2023-07-10 and effective 2023-08-15. The translation cited here is by China Law Translate, a crowdsourced legal translation project (Source: https://www.chinalawtranslate.com/en/generative-ai-interim/). The measure was issued by the Cyberspace Administration of China (CAC) together with six co-issuers: the NDRC, MoE, MoST, MIIT, MPS, and NRTA. See legislation page for structured provisions and comparisons.
Provenance
- Promulgators: Cyberspace Administration of China (CAC) with six co-issuers (NDRC, MoE, MoST, MIIT, MPS, NRTA)
- Promulgation date: 2023-07-10
- Effective date: 2023-08-15
- Translation: China Law Translate (crowdsourced legal translation project)
- URL: https://www.chinalawtranslate.com/en/generative-ai-interim/
Summary of provisions
The measure is organized into five chapters.
Chapter 1 (General Provisions) sets the scope to public-facing generative AI in mainland China, states a principle of "equal emphasis on development and security," establishes graded-category management, and imposes a socialist values requirement.
Chapter 2 (Development and Governance of Technology) encourages innovation and sets training-data requirements: lawful sources, respect for intellectual property, consent for use of personal information, and standards of quality, truth, accuracy, objectivity, and diversity. It also provides manual-tagging rules.
Chapter 3 (Service Specifications) assigns provider responsibility as information content producers and addresses user service agreements, disclosure obligations, confidentiality, a cross-reference to deep-synthesis labeling, and complaint mechanisms.
Chapter 4 (Oversight and Legal Responsibility) provides for multi-agency supervision. Under Article 17 it requires security assessments and algorithm filings for services with "public opinion properties or capacity for social mobilization." Article 20 sets a cross-border trigger, and penalties are graduated via existing cyber, data, and PIPL laws.
Chapter 5 (Supplementary Provisions) contains definitions and foreign-investment compliance terms.
Key claims
- CAC is the lead regulator but shares authority with six other bodies. (confidence: high)
- Services offered to the public trigger the rules; pure research and development does not. (confidence: high; Art. 2)
- Providers are liable as "producers of online information content," a content-moderation framing distinct from Western product-liability approaches. (confidence: high; Art. 9)
- Generative AI content must "Uphold Core Socialist Values" and must not generate content inciting subversion, separatism, terrorism, ethnic hatred, or "fake and harmful information." (confidence: high; Art. 4)
- Filing and security assessment are required for services with public-opinion or social-mobilization properties, via the algorithmic filing regime established under China — Internet Information Service Algorithmic Recommendation Management Provisions. (confidence: high; Art. 17)
- Deep-synthesis labeling is required via cross-reference to China — Provisions on the Administration of Deep Synthesis Internet Information Services. (confidence: high; Art. 12)
Context
The measure is the third and most capability-directed layer of China's regulatory stack, following the algorithmic recommendation provisions (2022) and the deep synthesis provisions (2023). It reuses the algorithmic filing regime from China — Internet Information Service Algorithmic Recommendation Management Provisions and the labeling regime from China — Provisions on the Administration of Deep Synthesis Internet Information Services.
It takes a different approach from EU AI Act (Regulation 2024/1689) (risk-tiered and horizontal), US AI Safety Institute — Vision, Mission, and Strategic Goals (voluntary and scientific), and the US sub-national Colorado AI Act (SB 24-205) and SB 25B-004 (Date Amendment) and California SB 53 — Transparency in Frontier AI Act. China's content-moderation framing makes it a distinct third regulatory track in global AI governance.
Relationships
- supports: legislation page
- depends-on: China — Internet Information Service Algorithmic Recommendation Management Provisions (filing regime reused)
- depends-on: China — Provisions on the Administration of Deep Synthesis Internet Information Services (labeling regime reused)
- related: EU AI Act (Regulation 2024/1689) (contrasting regulatory approach)
- related: US AI Safety Institute — Vision, Mission, and Strategic Goals (contrasting regulatory approach)
- related: AI and Authoritarianism
- related: AI and Content Moderation
- related: AI Sovereignty