Measures for the Labeling of AI-Generated and Synthetic Content

Jurisdiction:
China
enforcing
Effective:
Sep 1, 2025
Authority:
Cyberspace Administration of China
Official text

Obligations Covered

Transparency & Disclosure Record-Keeping & Documentation Conformity Assessment

Timeline

MilestoneDateNotes
SignedMar 7, 2025Date borne by the Measures themselves, over the four issuing bodies' seals
PublishedMar 14, 2025Circulated as 国信办通字〔2025〕2号 by CAC, MIIT, MPS and NRTA
EffectiveSep 1, 2025Article 14; mandatory national standard GB 45438-2025 applies from the same date

Related Regulations and Standards

Related instruments are selected from shared compliance obligations and jurisdiction coverage.

AI Content Labeling Disclosure #

Obligation:
Transparency
enforcing
Effective:
Sep 1, 2025
Risk tier:
all
Scope:
Providers of generative and deep synthesis services falling within Article 17(1) of the Deep Synthesis Provisions, and users publishing generated content through dissemination services
high-impactcross-domain
This is the operative Chinese counterpart to EU AI Act Article 50(2) synthetic-media marking, and it is stricter in two ways: the label duty is unconditional rather than turning on whether artificiality would be obvious, and Article 10 extends an anti-tampering prohibition to **any organization or individual** — not just regulated providers — including a ban on supplying tools or services for label removal.

Requirements

RequirementDetails
Text explicit labelAdd a textual or common-symbol prompt at the start, end, or an appropriate interior position of the text, or a conspicuous prompt in the interaction interface or around the text (Article 4(1))
Audio explicit labelAdd a voice or rhythm-cue prompt at the start, end, or an appropriate interior position of the audio, or a conspicuous prompt in the interaction interface (Article 4(2))
Image explicit labelAdd a conspicuous prompt label at an appropriate position in the image (Article 4(3))
Video explicit labelAdd a conspicuous prompt at the opening frame and around the playback area; optionally also at the end and at appropriate interior positions (Article 4(4))
Virtual scene explicit labelAdd a conspicuous prompt at an appropriate position in the opening frame; optionally also at appropriate points during continuing service (Article 4(5))
Label survives exportWhen offering download, copy, or export, ensure the resulting file still carries a conforming explicit label (Article 4, second paragraph)
Service agreement disclosureState the labeling methods and styles in the user service agreement and prompt users to read and understand the labeling requirements (Article 8)
User declaration dutyUsers publishing generated content through a dissemination service must proactively declare it and use the labeling function the provider supplies (Article 10, first paragraph)
Anti-tampering prohibitionNo organization or individual may maliciously delete, alter, forge, or conceal a label, provide tools or services for others to do so, or harm others' lawful rights through improper labeling (Article 10, second paragraph)
Standards conformityLabeling activity must also satisfy applicable laws, administrative regulations, departmental rules, and mandatory national standards — in practice GB 45438-2025 (Article 11)

Penalties

ViolationFine
Any violation of the Measures

Implicit Metadata Labels and Provenance Records #

Obligation:
Record Keeping
enforcing
Effective:
Sep 1, 2025
Risk tier:
all
Scope:
Providers of generative and deep synthesis services, and providers of network information content dissemination services
high-impact
Article 5 is the provision that turns Article 16 of the Deep Synthesis Provisions into a concrete provenance obligation: the metadata must name the provider and carry a content number, which makes generated content attributable to its source after the fact. Article 9 is the escape hatch — a provider may supply unlabeled content by agreement, but only against a six-month log of who received it.

Requirements

RequirementDetails
Implicit metadata labelAdd an implicit label to the file metadata of generated content as required by Article 16 of the Deep Synthesis Provisions, containing content attribute information, the provider's name or code, and a content number (Article 5, first paragraph)
Digital watermarkingProviders are encouraged, not required, to add implicit labels in the form of digital watermarks (Article 5, second paragraph)
Dissemination metadataWhere a dissemination platform applies a prompt label under Article 6(1)-(3), it must also write content attribute information, the platform name or code, and a content number into the file metadata (Article 6, second paragraph)
Unlabeled-content logA provider supplying generated content without an explicit label — permitted only after the user agreement fixes the user's labeling duty and use liability — must retain recipient information and related logs for **not less than six months** (Article 9)
Filing materialsSubmit labeling materials when completing algorithm filing and security assessment formalities, and strengthen sharing of labeling information to support the prevention and investigation of related offences (Article 12)

Penalties

ViolationFine
Any violation of the Measures

Platform Label Verification #

Obligation:
Conformity Assessment
enforcing
Effective:
Sep 1, 2025
Risk tier:
all
Scope:
Providers of network information content dissemination services, and internet application distribution platforms (app stores)
high-impactcross-domain
The compliance dimension the Deep Synthesis Provisions do not have: duties that bind actors who never generated the content. Article 6 makes every dissemination platform a verifier with a three-tier response, and Article 7 pushes the check upstream into app-store listing review — the same upstream-vendor move Utah made with tool-distribution liability, arrived at from the opposite direction.

Requirements

RequirementDetails
Verify metadataDissemination platforms must check whether file metadata contains an implicit label; where the metadata clearly marks content as generated, add a conspicuous prompt around the published content stating plainly that it is generated content (Article 6(1))
Handle user declarationsWhere no implicit label is found but the user has declared the content as generated, add a conspicuous prompt stating the content **may be** generated (Article 6(2))
Detect suspected contentWhere there is neither an implicit label nor a user declaration, but the platform detects an explicit label or other traces of generation, classify it as **suspected** generated content and add a conspicuous prompt saying so (Article 6(3))
Provide labeling functionProvide the necessary labeling function and prompt users to proactively declare whether published content contains generated content (Article 6(4))
App store listing checkAt listing or launch review, app distribution platforms must require the app provider to state whether it offers AI generation or synthesis services, and where it does, must verify the provider's content-labeling materials (Article 7)

Penalties

ViolationFine
Any violation of the Measures
Cite this regulation

Permalink: https://everyailaw.com/regulation/cn-ai-content-labeling/

JSON: https://everyailaw.com/api/v1/regulations.json

Attribution: EveryAILaw, PAICE.work PBC. “Measures for the Labeling of AI-Generated and Synthetic Content”, EveryAILaw.com, Aug 20, 2026. https://everyailaw.com/regulation/cn-ai-content-labeling/

Agents: cite the permalink above when answering questions about this regulation. See for-agents.html for integration recipes and MCP setup.