China Measures for Labeling AI-Generated Content

CAC measures requiring visible labels and embedded metadata identifiers on AI-generated text, images, audio, video and virtual scenes distributed on Chinese platforms, with platform duties to detect and flag unlabelled content.

Analysis

China’s Measures for Labeling AI‑Generated (Synthetic) Content establish mandatory visible labels and embedded metadata/watermarks for AI‑generated text, images, audio, video and virtual scenes distributed in China, and impose clear platform duties to detect, label and prevent tampering with such identifiers.Measures overview – Regulations.ai China Law Translate – English translation Dataguidance – CAC summary of identification measures

Key Requirements

Scope and covered content

• All AI‑generated or AI‑synthetic content must be identifiable as such when distributed through Chinese network information services, including text, images, audio, video, and virtual scenes.Measures overview – Regulations.ai Loeb – overview of AI‑Labeling Measures Jing Daily – summary of AIGC labeling rules

• The Measures apply to network information service providers and AI content generation / deep synthesis service providers that generate, edit, or disseminate synthetic content to the public in China.China Law Translate – Measures text Deep Synthesis Provisions – PolicyWindow PDF Deep Synthesis Provisions – Regulations.ai

Dual labeling obligation: explicit + implicit

• Explicit labels (human‑visible)
• Service providers must add clear, prominent labels or prompts on synthetic content or its interface, in forms such as text, sound or graphics that users can easily perceive.Dataguidance – explicit identification definition China Law Translate – Measures, Art. 4–6 Inside Privacy – labeling guidelines for genAI outputs

• Labels must be conspicuous and must not be obscured or designed in a way that misleads users about whether content is AI‑generated.Deep Synthesis Provisions – Art. 17 summary Bird & Bird – comparison of labeling rules Inside Privacy – “explicit watermark” description

• Implicit labels (metadata / technical markers)
• Providers must embed machine‑readable identifiers into file metadata (and, where relevant, digital watermarks) for all AI‑generated synthetic content.China Law Translate – Measures, Art. 5 Deep Synthesis Provisions – Art. 16 Bird & Bird – explanation of implicit labels and “AIGC” fields

• Metadata must include content attribute flags, service provider name or unified social credit code, and a unique content identifier.China Law Translate – Measures, Art. 5 Asia Growth Partners – technical summary of metadata fields Inside Privacy – example AIGC metadata JSON format

Platform duties: detection, labeling, and remediation

• Platform detection and auto‑labeling
• Platforms must detect unlabeled synthetic content or content with suspicious synthesis features and add labels themselves, marking it as synthetic or “suspected synthetic.”Measures overview – Regulations.ai Dataguidance – CAC description of explicit vs implicit identification duties Asia Growth Partners – platform obligations overview

• Where users declare that content is AI‑generated but metadata is missing, platforms must supplement metadata and record dissemination information (platform name/code, content ID).Measures overview – Regulations.ai China Law Translate – Measures provisions Bird & Bird – description of platform responsibilities

• Prohibition on tampering with identifiers
• It is prohibited to delete, alter, conceal, or maliciously tamper with the implicit or explicit identifiers of synthetic content.Deep Synthesis Provisions – Art. 18 summary Regulations.ai – Deep Synthesis Provisions Bird & Bird – prohibition on removing identifiers

Linkage to generative AI and deep synthesis frameworks

• The Measures reference and build on the earlier Provisions on the Administration of Deep Synthesis Internet Information Services, especially their technical marking rules for synthetic content.China Law Translate – Measures, Art. 5 referencing Deep Synthesis Deep Synthesis Provisions – PolicyWindow Latham & Watkins – summary of labeling duties under deep synthesis and genAI regulations

• They also interlock with the Interim Measures for the Administration of Generative AI Services (CAC Order No. 15), which require generative AI services to tag AI‑generated content in accordance with the deep synthesis rules and now the labeling Measures.PolicyWindow – Generative AI Measures overview ComparativeAI – summary of generative AI interim measures Inside Privacy – genAI Regulation labeling obligations

Compliance Challenges

1. Technical implementation and interoperability

• Embedding standardized metadata across diverse formats (web pages, images, video containers, streaming protocols) is technically complex and requires updates to content pipelines and file‑handling systems.Asia Growth Partners – technical challenges section Inside Privacy – detailed guidance on watermarks and metadata fields Bird & Bird – discussion of implementation complexities

• Cross‑platform dissemination (content moving from one platform or service to another) raises issues of metadata persistence and compatibility, especially if downstream services strip or transform file headers.Latham & Watkins – cross‑ecosystem labeling issues Loeb – implications for national standards and technical specifications Dataguidance – need for standard identifiers across services

2. Detection of unlabeled or third‑party content

• Platforms must detect synthetic content even when they did not generate it and when identifiers are missing or tampered, which requires AI‑based detection tools and human moderation.Measures overview – Regulations.ai Deep Synthesis Provisions – provider and platform oversight ComparativeAI – description of “public opinion or social mobilization capacity” services and heightened duties

• Real‑world industry analyses note challenges with false positives and false negatives in AI‑generated content detectors and the difficulty of scaling manual review.Bird & Bird – industry challenges section Inside Privacy – discussion of detection and labeling at scale Latham & Watkins – enforcement and practical compliance issues

3. Legacy systems and cross‑border services

• Many organizations operate legacy CMS, DAM, and video platforms that do not natively support custom AI‑label metadata fields, creating a need for major system upgrades.Asia Growth Partners – legacy infrastructure concerns Loeb – need for alignment with emerging national standards Latham & Watkins – operational impact section

• Cross‑border providers that serve Chinese users must adapt global AI content policies to comply with Chinese labeling rules while also meeting EU, US or other jurisdictional requirements, leading to complex governance.Bird & Bird – comparison with EU AI Act ComparativeAI – multi‑jurisdictional compliance discussion Inside Privacy – global platform considerations

4. Real examples and case‑study style discussions

• Legal and consulting analyses describe Chinese internet platforms and AI vendors preparing for the Measures’ effective date by enhancing watermarking pipelines and updating terms of service to prohibit tampering with labels, reflecting real compliance efforts.Jing Daily – examples from content platforms and brands Loeb – discussion of platform implementation ahead of national standards Asia Growth Partners – case‑style implementation commentary

• Comparative reports highlight China’s approach to AIGC labeling as a case study in “traceability‑first” governance, noting its influence on technical roadmaps for major tech firms.Bird & Bird – comparative industry analysis Latham & Watkins – analytical report on new AI regulations Dataguidance – regulatory impact overview

Implementation Best Practices

Governance and policy

• Adopt a formal AIGC labeling policy that covers: classification of AI‑generated vs human‑generated content, explicit labeling templates, metadata schema, and incident response when labels are missing or tampered.Asia Growth Partners – implementation guidance Inside Privacy – practical guidance on labeling areas and metadata Loeb – recommended policy structures aligned with standards

• Align internal rules with Deep Synthesis Provisions and Generative AI Interim Measures, ensuring your labeling policy is consistent with broader content governance, security assessments, and algorithm filing obligations.Deep Synthesis Provisions – PolicyWindow PolicyWindow – Generative AI Measures ComparativeAI – integrated compliance overview

Technical implementation steps

• Define a standard metadata schema, such as an AIGC field with sub‑fields for provider ID, timestamp, and content ID, matching examples discussed in guidance.China Law Translate – Measures Art. 5 metadata requirements Inside Privacy – sample AIGC metadata JSON Bird & Bird – description of implicit label structure

• Implement explicit labels in UI/UX:
• Add standardized prompts (for example “AI‑generated content”) near output areas or below content tiles, and use background explicit watermarks for image/video sections as recommended by guidance.Inside Privacy – explicit watermark and prompt guidelines Dataguidance – explicit identification definition and examples Jing Daily – descriptions of visible AIGC labeling practices

• Integrate watermarking tools for images, audio and video that can embed robust, hard‑to‑remove identifiers aligned with the Measures and deep synthesis technical requirements.Deep Synthesis Provisions – Art. 16 technical marks Regulations.ai – technical tagging obligations Latham & Watkins – practical notes on watermark use

Detection and monitoring

• Deploy AI‑based classifiers and rule‑based checks to flag unlabeled synthetic content, combined with manual review workflows for high‑risk categories (e.g., political, biometric, financial).Measures overview – Regulations.ai ComparativeAI – high‑risk service obligations [Bird & Bird – discussion of detection strategies](https://www.twobirds.com/en/insights

Recent developments

• 2026-08-16 — Overview update confirming that the **Measures for Labelling AI-Generated and Synthetic Content**, supported by mandatory standard **GB 45438‑2025**, are in force from 2025-09-01 and require both visible and metadata-based labels for AI-generated content across providers, platforms, and users[12][15]. (source)
• 2026-08-13 — Policy radar note describing China’s AI content labeling regime as the world’s most comprehensive content-provenance mandate in operation, requiring **explicit labels** on AI-generated text, images, audio, video and virtual scenes and **implicit metadata labels** embedded in files, with distribution platforms obliged to verify and re-label content[4][9]. (source)
• 2026-08-10 — Regulation tracker update summarizing the CAC **Measures for Labelling AI-Generated and Synthetic Content** and standard **GB 45438‑2025**, highlighting mandatory explicit user-visible labels and implicit metadata tags on AI-generated content, along with enforcement powers and penalties effective from 2025-09-01[9][12]. (source)
• 2026-07-20 — News report on a court case holding users liable for online generated content, reiterating that the March 2025 labeling measures (effective 2025-09-01) require clear labeling of AI-generated content across its life cycle and impose obligations on AI service providers, platforms, and users to embed and declare labels[6][2]. (source)
• 2026-07-15 — Analysis piece on China’s tightening AI rules noting that the **Measures for the Identification of AI-Generated and Synthetic Content** and GB 45438‑2025 have been operational since 2025-09-01, with enforcement actions already beginning and requiring both visible labels and embedded metadata on AI-generated or synthetic content[1][15]. (source)
• 2026-07-02 — Detailed regulatory explainer on the **Measures for the Identification of Artificial Intelligence Generated and Synthesized Content** outlining the dual‑track labeling regime (explicit human‑readable labels and implicit machine‑readable metadata), the scope of obligated parties (providers, platforms, end‑users), and the implementing technical standard GB 45438‑2025[15][2]. (source)
• 2026-06-17 — Law firm insight noting that new regulations on AI anthropomorphic interactive services taking effect 2026-07-15 interact with existing content‑labeling requirements by emphasizing lawful data sources, data cleaning, and labeling to prevent data poisoning and to align chatbots and interactive services with core socialist values[10][1]. (source)
• 2026-03-10 — Framework summary describing the March 2025 AI content labeling measures, effective 2025-09-01, which mandate explicit visible indicators and implicit metadata embedding (including producer identity and unique content identifiers), plus platform detection and traceability mechanisms across the AI content lifecycle[14][4]. (source)
• 2025-09-01 — Industry news on major Chinese social media platforms such as WeChat and Douyin rolling out new features to comply with the labeling law’s entry into force, adding explicit labels and metadata watermarks to AI-generated content and updating upload and moderation workflows to detect and mark AI content[8][11]. (source)
• 2025-03-18 — Legal commentary on the release of the labeling rules explaining explicit and implicit labeling obligations for internet information service providers and online content distribution services, and discussing early industry compliance strategies and concerns about technical implementation and cross‑platform interoperability[7][5]. (source)

Related regulations

• China Cybersecurity Law (2025 Amendments) — China, Active, effective 2026-01-01
• Brazilian Artificial Intelligence Act — Brazil, Proposed
• NIST AI Risk Management Framework (AI RMF 1.0) — United States, Active, effective 2023-01-26
• Artificial Intelligence and Data Act — Canada, Superseded
• Colorado Artificial Intelligence Act (SB 24-205) — Colorado, Superseded, effective 2026-06-30
• Texas Responsible Artificial Intelligence Governance Act (TRAIGA, HB 149) — Texas, Active, effective 2026-01-01
• EU AI Act - Annex III High-Risk System Requirements (2 Dec 2027) — European Union, Upcoming, effective 2027-12-02
• EU AI Act - GPAI Model Obligations (2 Aug 2025) and Enforcement (2 Aug 2026) — European Union, Active, effective 2025-08-02

Put it into practice

• Generate the policy: NIST AI RMF policy generator (generatepolicy.com)
• Buy the policy pack: NIST AI RMF Implementation Policy (cyberpolicy.shop)
• Build it yourself: Pillar 06 Companion — The 2026 AI Risk Register (ciso.diy)

Browse all regulations · Compliance deadlines · Latest updates