Skip to content

Provisions on the Administration of Deep Synthesis Internet Information Services

📑 Legal hierarchy: Level 3 · Departmental rule | Issuance: CAC, MIIT, MPS (three-ministry joint issuance) | Effective: 2023-01-10 | Character: hard law

⚠️ Hierarchy note: This instrument is a departmental rule, jointly issued by three ministries. It is not a State Council administrative regulation. Its penalty provisions route back to upstream laws such as the Cybersecurity Law. See Index of Chinese Rules.

English Summary

Section titled “English Summary”

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》), jointly issued by the Cyberspace Administration of China (CAC), the Ministry of Industry and Information Technology (MIIT), and the Ministry of Public Security (MPS) on 2022-12-11 and effective 2023-01-10, is China’s first dedicated regulation on deep synthesis technologies — covering face/voice manipulation, generative text, and virtual scenes. The Provisions establish a tiered obligation regime for service providers and technical supporters, introduce algorithm filing requirements for services with “public opinion attributes or social mobilization capacity,” and mandate conspicuous labeling of synthetic content that may confuse the public. These Provisions are the upstream rule from which the 2025 Labeling Measures directly derive their labeling obligations (see biaozhi-banfa).

Overview

Section titled “Overview”

The Deep Synthesis Provisions (2023) were China’s pre-emptive rule before generative AI, with a prologue reaching back to 2019 when the State Council Information Office jointly issued documents targeting deepfakes in “online audio and video information services.” After a draft for public comment in January 2022 and finalization in December 2022, they took effect on 2023-01-10.

The Provisions’ historical significance has three layers:

  1. They were the first to treat “deep synthesis” as an independent regulatory object, no longer dispersed across generic rules such as the Provisions on the Governance of the Online Information Content Ecosystem.
  2. They were the first to require “conspicuous labeling” of synthetic content (Article 17) — though they did not specify how to label; this gap was later filled by the 2025 Labeling Measures.
  3. They established “algorithmic filing (备案)” as a compliance gateway, subsequently inherited by the Generative AI Measures and the Labeling Measures as a shared anchor.

Scope

Section titled “Scope”
  • Subjects:
    • Service providers: platforms / apps that directly offer deep-synthesis functionality.
    • Technical supporters: third parties that provide deep-synthesis technology to service providers.
  • Activities:
    • editing and synthesis of biometric information such as faces and voiceprints;
    • generation and composition of text, images, audio, video, and virtual scenes;
    • generation of immersive realistic scenes.
  • Territory: services provided within China.

Core Duties

Section titled “Core Duties”

The following is this site’s summary; article numbers follow the official text.

On service providers

Section titled “On service providers”
  • Real-name verification: verify the real-identity information of users of deep-synthesis services.
  • Conspicuous labeling: where content may cause public confusion or misidentification, add a conspicuous label at a reasonable position or region of the generated content (Article 17 — the upstream basis of the 2025 Labeling Measures).
  • Algorithm filing (备案): services with public-opinion attributes or social-mobilization capacity shall file under the Algorithm Recommendation Provisions (see algorithm-recommendation-provisions).
  • Processing biometric information: when editing biometric information such as faces or voices (e.g., face-swaps, voice-cloning), providers shall notify users and obtain separate consent from the person being edited.
  • Data and model security: compliance of training data, security assessment of models, preservation of logs.

On technical supporters

Section titled “On technical supporters”
  • Share the duties on data and model security.
  • Assist service providers in fulfilling the conspicuous-labeling duty.

On users

Section titled “On users”
  • Shall not use deep-synthesis services for activities prohibited by law.

Enforcement and Penalties

Section titled “Enforcement and Penalties”
  • Competent authorities: CAC leads; MIIT (compute, telecommunications) and MPS (public security and criminal matters) coordinate within their mandates.
  • Penalty basis: the existing ladders under the Cybersecurity Law, Data Security Law, Personal Information Protection Law, and Public Security Administration Punishments Law.
  • Serious violations: criminal liability applies.

Relationship with Other Rules

Section titled “Relationship with Other Rules”
  • Algorithm Recommendation Provisions (2022): the filing obligation is incorporated by reference through Article 19 of these Provisions; the two share the same “algorithmic filing” infrastructure.
  • Generative AI Interim Measures (2023): lift generative AI services from a mere subset of “deep synthesis technology” to an independent regulatory object; the duty sets overlap with these Provisions but are more granular.
  • Measures for the Identification of AI-Generated and Synthesized Content (2025): the technical landing of Article 17’s “conspicuous labeling” requirement — see biaozhi-banfa.
  • Personal Information Protection Law (2021): applies jointly where faces, voiceprints, or other sensitive information are processed; these Provisions are in effect the specialized implementing rules of PIPL Article 28 in the deep-synthesis scenario.

Debates and Commentary

Section titled “Debates and Commentary”
  1. Subjectivity of the “may cause public confusion or misidentification” threshold: Article 17 conditions the duty to label on a “may-confuse-or-mislead” test. How is this to be determined in practice? Do fully photorealistic face-swaps count? What about satire? The Labeling Measures effectively lower the threshold to zero (label everything), generating some tension with the original intent of these Provisions.
  2. Practical threshold of algorithmic filing (备案): by the end of 2024, over a thousand algorithms had been filed. The process looks procedural, but deletion or modification of filings faces significant friction — it has become a de facto licensing gate.
  3. Cross-border applicability: enforcement against overseas deep-synthesis services accessible within China remains limited.
  4. Realism of the technical supporter duty: holding upstream model providers responsible for downstream service providers’ compliance creates boundary-drawing problems in practice.

Source Text and Translations

Section titled “Source Text and Translations”
LanguageSourceLink
Chinese (original)CACcac.gov.cn
Chinese (archived copy)This sitedeep-synthesis-provisions-2022-12-11.html
EnglishChina Law Translate (Jeremy Daum, Yale Paul Tsai China Center)chinalawtranslate.com/en/deep-synthesis
English (draft for comment, with analysis)Stanford DigiChinadigichina.stanford.edu/work/…-jan-2022
English (practitioner commentary)Allen & Gledhillallenandgledhill.com/…/22947

Version History

Section titled “Version History”
DateEvent
2022-01-28Draft for public comment released
2022-12-11Official release
2023-01-10Effective
2026-04-21First archived on this site

Cited Sources

Section titled “Cited Sources”

Primary

Section titled “Primary”

Major commentary

Section titled “Major commentary”

Cite this page (generated 2026-04-21):

Comparative AI. Commentary on the Provisions on the Administration of Deep Synthesis Internet Information Services. Accessed YYYY-MM-DD. https://comparativeai.org/rules/china/deep-synthesis-provisions/