Skip to content

China — Content Labeling and Provenance

This page is thematic analysis. For the actual legal text of each instrument, see the corresponding rule pages; this page only synthesises the analysis and observations across rules.

Relevant rules

Section titled “Relevant rules”
RuleLegal tierRule page on this site
Measures for the Identification of AI-Generated Synthetic Content 《人工智能生成合成内容标识办法》(2025)Departmental rule · hard lawbiaozhi-banfa
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》(2023)Departmental rule · hard lawdeep-synthesis-provisions
Interim Measures for the Management of Generative Artificial Intelligence Services 《生成式人工智能服务管理暂行办法》(2023)Departmental rule · hard lawgenerative-ai-interim-measures
Provisions on the Administration of Algorithmic Recommendation of Internet Information Services 《互联网信息服务算法推荐管理规定》(2022)Departmental rule · hard lawalgorithm-recommendation-provisions
GB 45438-2025Mandatory national standard · soft law*gb-45438-2025-ai-content-labeling-standard

* See methodology §2 for classification rationale.

The three-phase evolution of China’s labeling regime

Section titled “The three-phase evolution of China’s labeling regime”

This is the analytic spine of this page — a three-step reading of how China’s policy on AI-synthetic-content labeling evolved.

Phase 1 (2017–2022): scattered “shall be conspicuously labelled” clauses

Section titled “Phase 1 (2017–2022): scattered “shall be conspicuously labelled” clauses”
  • The Provisions on the Governance of the Online Information Content Ecosystem, the Provisions on the Administration of Internet Live Streaming Services, and others made occasional references to “conspicuous labeling”.
  • No unified terminology, no technical standard, low enforceability.

Phase 2 (2023): two specialised rules establish principles — deep synthesis and generative AI

Section titled “Phase 2 (2023): two specialised rules establish principles — deep synthesis and generative AI”
  • Deep Synthesis Provisions: for the first time articulated a service provider’s conspicuous-labeling obligation (art. 17).
  • Generative AI Interim Measures: layered in filing, security assessment, and training-data compliance requirements.
  • Open questions: how exactly should a label be applied? what are the metadata fields? how do platforms verify? — no answers.

Phase 3 (2025): the Labeling Measures + GB 45438 translate policy into technical implementation

Section titled “Phase 3 (2025): the Labeling Measures + GB 45438 translate policy into technical implementation”
  • The Labeling Measures introduce a dual-track explicit + implicit regime.
  • GB 45438-2025 defines interoperable fields.
  • Obligor scope expands from service providers to a three-way service provider + platform + user allocation.

Differences from other jurisdictions (thematic-level analysis)

Section titled “Differences from other jurisdictions (thematic-level analysis)”

This is where the topic page adds value. This section does not restate legal text — it offers comparative analysis.

  • Versus EU AI Act art. 50: the EU contains an explicit art / satire exemption for deepfakes; China’s Labeling Measures has no express exemption, so how “artistic creation” will be recognised in practice is unclear.
  • Versus US state law: US state law centres on elections + non-consensual synthesis, and China’s coverage is broader than the US’s but narrower than the EU’s (no risk-tier obligations).
  • Ex-ante binding to a technical standard is a distinctive Chinese feature: GB 45438 enters into force on the same day as the Measures, forming a “departmental rule + national standard” one-two combination; the US NIST counterpart remains at the level of voluntary frameworks.