This page summarizes the China Law Translate English translation of the Provisions on the Administration of Deep Synthesis Internet Information Services, China's regulation governing "deep synthesis" (generative and synthetic) internet information services. The Provisions were jointly issued on November 25, 2022 and took effect January 10, 2023. For the structured provisions, see the legislation page.
Provenance
- Promulgators: CAC, MIIT, and the Ministry of Public Security
- Promulgation date: 2022-11-25
- Effective date: 2023-01-10
- Translation: China Law Translate
- URL: https://www.chinalawtranslate.com/en/deep-synthesis/
Summary of provisions
The Provisions are organized into four chapters.
Chapter I (General) defines "deep synthesis" as the use of deep-learning, virtual-reality, and other generative-sequencing algorithms to produce text, audio, image, video, 3D, or immersive-scene content. It sets the scope of the rules and provides for multi-agency coordination.
Chapter II (Ordinary Provisions) prohibits illegal content and bars the use of deep synthesis to produce fake news. It places primary-responsibility obligations on providers, including registration, algorithm review, data security, ethics review, fraud prevention, and emergency response. It requires real-name verification of users (Art. 9), content review combining technical and manual methods, a rumor-dispelling mechanism, and app-store gatekeeping (Art. 13).
Chapter III (Data and Technology) governs training-data management. It requires consent when biometric information such as faces or voices is edited (Art. 14) and security assessments for biometric editing and sensitive content (Art. 15). It establishes a labeling regime consisting of log-level labels (Art. 16) and conspicuous public-facing labels for services likely to confuse or mislead users (Art. 17, covering smart dialogue and writing, voice synthesis, face generation and swap, gesture manipulation, and immersive scenes). Tampering with labels is prohibited (Art. 18).
Chapter IV (Oversight) requires filing through the algorithmic filing regime for providers with public-opinion or social-mobilization capacity (Art. 19), security assessments for new products and functions (Art. 20), and enhanced penalties where serious consequences result.
Key claims
The source summary records the following claims, each rated high confidence:
- The Provisions establish what is described as the first comprehensive national deepfake and synthetic-media labeling regime globally.
- They require real-name verification for deep-synthesis service users (Art. 9), a feature the summary identifies as distinctive.
- They mandate consent from individuals whose biometric data (face or voice) is edited (Art. 14).
- They create a dual-tier labeling regime combining technical labels (log or watermark) with public-visible labels (Arts. 16–17).
- They reuse the filing and security-assessment regime from the 2021 algorithmic recommendation rules (Art. 19).
- They extend obligations to "technical supporters" (upstream model providers), not only deployers (Art. 3, Art. 14).
Place in China's regulatory stack
The Provisions sit as the middle layer of China's AI regulatory stack, between general algorithmic governance (2022) and capability-specific generative-AI rules (2023). The labeling regime predates the transparency obligations in EU AI Act (Regulation 2024/1689) Article 50 and offers an earlier working precedent. The real-name verification requirement has no direct Western analogue and reflects the broader Chinese cyber identity framework.
Relationships
- supports: legislation page
- depends-on: China — Internet Information Service Algorithmic Recommendation Management Provisions (reuses filing regime)
- related: China — Interim Measures for the Management of Generative AI Services (reuses labeling regime via cross-reference)
- related: AI and Content Moderation
- related: AI and Authoritarianism
- related: EU AI Act (Regulation 2024/1689) (transparency/labeling comparison)
- related: AI Sovereignty