Introduction
China's AI regulatory regime is not a single statute but a stack of instruments issued in quick succession, layered on top of the data and cybersecurity framework. A foreign company deploying an AI system in China — or serving Chinese users from outside — has to satisfy obligations that are easy to miss until the filing deadline has passed. This article maps the three core instruments, the generative-AI filing and labeling duties, and the state of the long-discussed AI law.
The Three Core Instruments
Three CAC-led instruments form the base layer, all currently in force:
- 《互联网信息服务算法推荐管理规定》 (Provisions on the Administration of Algorithmic Recommendations) — issued by four departments, CAC Order No. 9, effective 1 March 2022.
- 《互联网信息服务深度合成管理规定》 (Provisions on the Administration of Deep Synthesis) — issued by three departments, CAC Order No. 12, effective 10 January 2023.
- 《生成式人工智能服务管理暂行办法》 (Interim Measures for the Administration of Generative AI Services) — issued by seven departments, CAC Order No. 15, effective 15 August 2023.
The Generative-AI Obligations That Matter Most
For a company offering generative AI, the Interim Measures carry two obligations that drive most compliance work. Article 17 requires a provider whose service has "public-opinion attributes or social mobilization capability" (具有舆论属性或者社会动员能力) to conduct a security assessment and complete the algorithm filing under the algorithm-recommendation rules — the so-called "large-model filing" route, layered on top of ordinary algorithm filing. Article 12 requires providers to label generated content in accordance with the deep-synthesis rules. Article 16 allocates supervision: CAC leads, with the NDRC, Ministry of Education, MOST, MIIT, MPS and NRTA each supervising within their remit.
Labeling Has Moved to a 2025 Measure
An article citing only the 2023 rules understates the current labeling regime. The dedicated 《人工智能生成合成内容标识办法》 (Measures for the Labeling of AI-Generated and Synthetic Content), issued by four departments and effective 1 September 2025, now governs both explicit and implicit labeling, and is paired with the mandatory national standard GB 45438-2025 (same effective date). Compliance programs should cite the 2025 measure for labeling, not the 2023 deep-synthesis provision alone.
Is There an AI Law?
Not yet. No standalone Artificial Intelligence Law has been enacted, and no draft is currently under review by the NPC Standing Committee. The State Council's 2025 legislative work plan (published 14 May 2025) changed the formulation to "推进人工智能健康发展立法工作" — advancing legislation for the healthy development of AI — which signals continued drafting rather than imminent enactment. This is the single most time-sensitive point: track the legislative status against the latest official announcements.
A Practical Compliance Checklist
- Classify the service — public-facing generative AI, an algorithm-recommendation product, or an internal tool; each maps to different instruments.
- Test the "public-opinion attributes" trigger — if met, plan for security assessment plus algorithm filing under Article 17.
- Implement labeling — explicit and implicit, per the 2025 labeling measure and GB 45438-2025.
- Confirm the supervisory remit — CAC leads, but sector regulators apply in their lanes.
- Watch the draft AI law — obligations can change when it is tabled.
For the data and cybersecurity framework this regime sits on, see China Cybersecurity and PIPL Enforcement: A Defence Guide, or contact us to map the filing obligations for a specific product.
How Tianni Law Firm Can Help
Tianni Law Firm (江苏天倪律师事务所), based in Nanjing, Jiangsu, advises foreign companies on technology regulation and compliance in China. Our work includes mapping AI and algorithm-filing obligations, security-assessment preparation, content-labeling and data-compliance programs, and representation before the CAC and sector regulators. This article is general information, not legal advice; confirm the operative instruments and any pending legislation before relying on it for a filing.
Deploying AI or an algorithm product in China?
We map your filing, security-assessment and labeling obligations before launch — and track the draft AI law so you are not caught mid-release.
Get a Free Consultation