China's Generative AI Measures: the Registration and Filing Duty
In mainland China, a generative AI service aimed at the public cannot simply be switched on. Two separate administrative steps sit between a working product and a lawful public launch, and they are governed by two different instruments that are routinely conflated. The instrument and its date The Interim Measures for the Management of Generative AI Services (生成式人工智能服务管理暂行办法) were issued on 13…
In mainland China, obtaining a public launch of a generative AI service is not a straightforward process. The service must first navigate two distinct administrative steps governed by the Interim Measures for the Management of Generative AI Services, issued by the Cyberspace Administration of China on July 13, 2023. The term "interim" in the title signifies that this is a provisional instrument, subject to change through subsequent regulations, such as the 2025 content labelling measures.
The scope of these measures encompasses any use of generative AI technology to produce services that create text, images, audio, video, or similar content for the public within mainland China. However, two important exceptions exist. First, research and internal development activities that do not expose the generative AI to the domestic public are exempt. Second, a service that does not target or serve mainland users, even if developed domestically, falls outside the purview of these measures.
The filing duty, outlined in Article 17, is more narrowly focused than the Measures themselves. Providers of generative AI services that possess public opinion attributes or social mobilization capabilities must undergo a security assessment and submit an algorithm filing. This phrase captures a broad range of services, including those that allow public users to generate and view content or organize around it.
Conversely, a business-to-business tool with no public-facing content surface may not be subject to this requirement, though this determination requires a case-by-case evaluation.
The required sequence of actions is non-negotiable. First, ascertain whether the service possesses the characteristics of public opinion attributes or social mobilization capability. If it does not, the provider is exempt from the filing obligation and still needs to comply with the security assessment outlined in the 2018 Provisions on the Security Assessment of Internet Information Services.
Subsequently, the provider must complete the algorithm filing through the CAC's filing system. Once the service is publicly launched, the filing must be kept current. Any material changes to the algorithm or cessation of the service will necessitate additional filings within specified timeframes.
Article 17 serves as a precondition for public operation, not a mere formality. The CAC maintains a public list of generative AI services that have completed the filing process, making it uncommon for a domestic consumer launch to go unnoticed if the provider's name is absent from this list. The filing itself demands a description of the algorithm and related accountability measures rather than a technical audit.
It necessitates the provider's identification, the service's format and domain, the type of algorithm used, and a self-assessment report detailing adherence to content requirements, training data usage, provider responsibilities, labelling rules, and procedures for handling complaints and correcting unlawful content. Foreign providers often make missteps by overlooking these comprehensive obligations.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — it may contain errors, so check the original before relying on it.