China Deep Synthesis Regulations

Applies To

Overview

The Provisions on the Administration of Deep Synthesis Internet Information Services (《互联网信息服务深度合成管理规定》) were promulgated jointly by the Cyberspace Administration of China (CAC), the Ministry of Industry and Information Technology, and the Ministry of Public Security. They came into effect on 10 January 2023. The provisions define 'deep synthesis' broadly to encompass any technology that uses deep learning (a core AI technique), virtual reality, or related techniques to generate or materially alter text, images, audio, video, virtual scenes, or virtual persons. This definition captures generative AI models, voice cloning, face swapping, lip-sync manipulation, text-to-image systems, AI-generated avatars, and similar capabilities. The regulations apply to deep synthesis service providers (entities that develop or operate platforms offering deep synthesis tools) and to deep synthesis users who deploy those tools to create or disseminate content. The framework sits within China's broader AI content governance architecture, which includes the Generative AI Measures (effective August 2023), the Algorithm Recommendation Provisions (effective March 2022), and the overarching Cybersecurity Law and Data Security Law. The provisions apply to deep synthesis services provided within China. Companies offering such features there should check whether the provisions cover their products and content supply chain. Particular attention is required for enterprises using third-party AI-generated content APIs (software connections that pull in content from outside vendors), synthetic media in advertising, AI avatars in customer service, and AI-generated dubbing or subtitling.

Key Requirements

• •Providers must verify users' real identities, and must not offer information publishing services to unverified users (Article 9).
• •Synthetic content that could cause public confusion must carry a visible and non-removable label disclosing its AI-generated or AI-modified nature; providers must embed imperceptible technical watermarks in addition to visible labels.
• •Providers must not be used to create or disseminate disinformation, fake news, content that endangers national security, content that defames individuals, or content that infringes on personal information or portrait rights.
• •Content generated using the likeness, voice, or biometric data of a real person requires the prior explicit consent of that individual.
• •Providers with public opinion attributes or social mobilization capability must file their algorithms (Article 19) and assess new features before launch (Article 20).
• •Tools that generate or edit faces, voices, or other biometric information need a security assessment (Article 15).
• •Providers must implement content-moderation systems capable of detecting and blocking prohibited deep synthesis outputs before distribution.
• •Training data used to build deep synthesis models must be sourced lawfully, respecting intellectual property rights and personal-information-protection obligations under PIPL and the Data Security Law.
• •Providers must establish and publish complaint and reporting mechanisms allowing users to flag suspected illegal or harmful synthetic content.
• •The provisions apply to services provided within China; they set no user thresholds or local representative duty for overseas providers.

What Your Organization Must Do

• →Map which products offered in China generate or materially alter text, images, audio, video or avatars.
• →Add visible, non-removable AI labels plus hidden watermarks to synthetic content before users see it.
• →Verify user identities by mobile number, ID number or similar, and block publishing by unverified users.
• →Collect prior explicit consent before using any real person's likeness, voice or biometric data.
• →File algorithms and assess new features where the service shapes public opinion or mobilises people.
• →Review third-party content feeds, AI dubbing, avatars and ad media for lawful training data and complaint channels.

Playbook Guidance

Step-by-step implementation guidance for compliance teams.

Governance Controls

Operational controls that implement requirements from this regulation.

Frequently Asked Questions

Do the China Deep Synthesis Regulations apply to foreign companies whose AI tools are accessible to users in China?

The provisions apply to deep synthesis services provided within China, with no user thresholds or local representative duty. Foreign companies serving users in China should take local advice, since other laws such as PIPL can apply.

What labelling is required for AI-generated content under the Deep Synthesis Provisions?

Providers must apply both a visible, non-removable label disclosing AI-generated or AI-modified status and an imperceptible technical watermark. Both controls must be embedded in the content pipeline before any synthetic output reaches end users.

Is real-name registration mandatory for users of deep synthesis platforms in China?

Yes, providers must verify users' identities by mobile number, ID number, or similar means. They must not offer information publishing services to users who have not been verified.

When does using a real person's likeness or voice in AI-generated content require consent under these provisions?

Any use of a real individual's likeness, voice, or biometric data in synthetic content requires prior explicit consent from that person. Consent workflows should be reviewed by legal counsel for compliance with PIPL and Chinese portrait-rights requirements.

What triggers the CAC security assessment and algorithm registration requirement under the Deep Synthesis Provisions?

Services with public opinion attributes or social mobilization capability must file their algorithms and assess new features before launch. Separately, tools that generate or edit faces, voices, or other biometric information need a security assessment.

How do the Deep Synthesis Provisions relate to China's Generative AI Measures and Algorithm Recommendation Provisions?

The Deep Synthesis Provisions form one layer of China's AI content governance framework, sitting alongside the Algorithm Recommendation Provisions (effective March 2022) and the Generative AI Measures (effective August 2023). Providers may face overlapping obligations under all three instruments depending on the functionality of their service.