China - AI Content Identification (RAI-CN-NA-MIASCXX-2025)

Overview

The Measures for the Identification of AI-Generated (Synthetic) Content ("the Measures") were jointly promulgated in March 2025 by the Cyberspace Administration of China (CAC), the Ministry of Industry and Information Technology (MIIT), the Ministry of Public Security (MPS) and the National Radio and Television Administration (NRTA). The Measures set a mandatory national baseline requiring that AI-generated or AI-synthesized content (including text, images, audio, video and virtual scenes) be clearly identified to users via two complementary mechanisms: explicit identification (visible labels and interface prompts) and implicit identification (machine-readable metadata or digital watermarks embedded in file headers). The Measures operate together with the mandatory national technical standard GB 45438-2025 and existing laws such as the Cybersecurity Law and sectoral rules on algorithm recommendation and deep synthesis. The official promulgation and notice can be found on the CAC website: Cyberspace Administration of China - Notice.

Definitions

The Measures define "AI-generated/synthetic content" broadly as information produced by artificial intelligence technologies — including text, images, audio, video and virtual scenes. "Explicit identification" refers to labels or prompts presented in human-perceptible form (text, audio cues, on-screen badges, graphics) added to the generated content or the user interface where it is presented. "Implicit identification" refers to technical markers embedded in the file metadata or using digital watermarks that are not obvious to end users but can be automatically detected by compliant platforms for provenance verification. The Measures also make a distinction between content-generation service providers (who create synthetic content) and information dissemination service providers (platforms that host or distribute that content).

Governance and Institutional Framework

The Measures assign primary supervising roles to central sectoral agencies. The policy text and implementing guidance were published jointly by the CAC together with MIIT, MPS and NRTA. The CAC is the primary coordinating authority for internet content policy and technical standards, while MIIT brings telecommunications and industrial oversight, MPS is responsible for criminal and public security enforcement where applicable, and NRTA supervises broadcast and audiovisual distribution. Relevant pages with official publication and Q&A are maintained by the CAC: CAC Q&A on the Measures, and MIIT also published a summary announcing the effective date: MIIT announcement. Enforcement is designed as a co-operative, multi-agency approach where administrative measures, sector supervision and law enforcement actions can be taken in accordance with existing statutory powers.

Key Focus Areas

The Measures focus on five principal areas: (1) clear, modality-specific explicit labeling that prevents user confusion across text, image, audio, video and virtual scenarios; (2) mandatory machine-readable metadata and encouraged digital watermarking to provide traceable provenance that persists when files are downloaded, exported or copied; (3) distributor and platform responsibilities to verify metadata, surface prominent prompts when synthetic content is identified (or suspected), and support user declarations; (4) app store and channel vetting to ensure applications offering generation services have labeling controls and compliance materials at the point of listing; and (5) safeguards against tampering and malicious tools that remove or alter labels. These areas are implemented through both regulatory obligations in the Measures and detailed technical prescriptions in the companion standard GB 45438-2025 (the national standard on labeling methods), which provides encoding rules, metadata field definitions and recommended watermarking approaches.

Implementation Framework

Operationally, service providers that generate synthetic content must embed implicit metadata into file headers containing key production and provenance attributes (e.g., content attribute flags, provider name or code, content ID) and apply explicit, human-visible labels according to the content format detailed in the Measures. Platforms that redistribute content must check for implicit markers; if metadata is present and denotes a synthetic origin, they must add visible prompts around the content. Where metadata is absent but the user declares synthetic origin or platforms detect explicit synthesis traces, platforms must similarly label the content as synthetic or suspected synthetic and add metadata fields that capture dissemination details (platform name/code, content ID). The Measures allow for controlled exceptions: providers may supply unlabelled content when a user requests that for legitimate social or industrial reasons, but providers must (a) obtain a clear agreement defining user labeling obligations and liabilities and (b) retain logs of recipient identification and service provision records for a minimum of six months. App distribution platforms are required to verify application labeling capabilities during launch or listing reviews.

Monitoring and Evaluation

Monitoring under the Measures is multi-layered: platforms must implement automated verification to detect implicit metadata and explicit labeling; agencies (CAC, MIIT and NRTA) will conduct compliance checks and audits; public security organs will investigate alleged criminal misuse; and industry bodies and standards committees will track technical performance of metadata/watermark schemes. The companion national standard GB 45438-2025 establishes technical validation norms that platforms can use to test compliance. Because the labeling approach relies on persistent file metadata, regulators and auditing tools can periodically sample content flows to measure labeling rates, incidence of tampering, and false negatives/positives in automated detection. The Measures also require providers to retain logs (including for cases where unlabeled content is delivered at user request) to enable post-event reconstruction and enforcement.

Penalties, Liability, and Appeals

The Measures state that violations will be handled by competent departments — including the CAC, telecommunications regulators, public security organs and radio and television authorities — in accordance with their statutory powers and relevant laws and regulations. While the Measures do not list specific monetary fines within the text itself, they expressly preserve administrative, civil and criminal routes available under Chinese law for unlawful conduct (for example, administrative sanctions for regulatory noncompliance, public security penalties for actions that breach public security regulations, and criminal liability where acts constitute crimes). The prohibitions include malicious deletion, alteration, forgery or concealment of labels and the provision of tools that facilitate such actions. Affected parties retain the right to administrative reconsideration and litigation under the Administrative Procedure Law and related frameworks where applicable; compliance guidance and dispute resolution processes are expected to be provided by implementing agencies through follow-on notices and judicial/administrative channels.

Relationship to Other Instruments

The Measures explicitly reference and operate in conjunction with the Cybersecurity Law, the Internet Information Service Algorithm Recommendation Regulation, the Internet Information Service Deep Synthesis Regulation, and the Interim Measures on Generative AI Services. A mandatory national standard (GB 45438-2025) was issued in parallel to provide technical methods for labeling and metadata encoding. Collectively, these instruments establish a layered governance architecture: high-level legal obligations (Cybersecurity Law and administrative rules), operational labeling obligations in the Measures, and technical specifications in the GB standard that implementers must follow. Service providers are required to comply with all applicable laws and standards and to submit labeling materials during algorithm filing and security assessment procedures to other regulatory processes referenced in these laws.

International Alignment

The Measures share objectives with international policy trends that call for provenance, transparency and traceability of synthetic media; however, the Chinese approach emphasizes mandatory labeling, persistent metadata, and state-supervised enforcement combined with sectoral technical standards. This places the Measures alongside other jurisdictions developing platform labeling requirements and technical provenance standards, but with distinct features: mandatory national technical standards (GB 45438-2025) and integration into algorithm filing and platform oversight regimes. Entities operating cross-border services should consider both the Measures' extraterritorial reach (where services are provided to users within China) and potential interactions with foreign provenance efforts, voluntary standards, and privacy/data transfer frameworks. Regulators have signalled willingness to coordinate with industry on implementation specifics while emphasising national security and public interest protections.

Implementation Timeline

Sources and References