Interim Measures for the Management of Generative AI Services — Comprehensive Review

Executive Summary

The "Interim Measures for the Management of Generative AI Services" (Order No. 15) was jointly issued by seven Chinese government departments — the Cyberspace Administration of China (CAC), the National Development and Reform Commission, the Ministry of Education, the Ministry of Science and Technology, the Ministry of Industry and Information Technology, the Ministry of Public Security, and the National Radio and Television Administration — on July 13, 2023, taking effect on August 15, 2023. It is one of the world's first comprehensive regulatory instruments specifically targeting generative AI services, marking the formal establishment of China's regulatory framework for large language models such as ChatGPT and their derivative services.

The measures contain 24 articles across five chapters, grounded in higher-level legislation including the Cybersecurity Law, the Data Security Law, the Personal Information Protection Law, and the Science and Technology Progress Law. They establish the fundamental principle of "balancing development and security, combining innovation promotion with governance according to law," and explicitly mandate "inclusive and prudent, classified and tiered regulation" of generative AI services. The scope covers services that use generative AI technology to provide text, image, audio, video, and other content generation to the public within China's borders. Critically, the measures exclude pure technology R&D activities that do not provide services to the public, reflecting a deliberate protection of innovation.

Key Provisions

The core provisions can be organized into four major categories:

1. Content Compliance and Values Requirements (Article 4): This is the most scrutinized provision. Providers and users must comply with laws, respect social morality and ethics. Specifically: adherence to core socialist values; prohibition of generating content that endangers national security, damages national image, incites separatism, promotes terrorism or extremism, or disseminates false and harmful information; prevention of discrimination based on ethnicity, religion, nationality, geography, gender, age, profession, or health throughout algorithm design, training data selection, model generation, and optimization; respect for intellectual property and commercial ethics; protection of others' lawful rights; and enhancing transparency and content accuracy based on service type characteristics.

2. Training Data and Technical Governance (Articles 7-8): Providers must use data and base models with lawful sources; must not infringe intellectual property rights; must obtain personal consent or meet statutory conditions when personal information is involved; and must take effective measures to improve training data quality in terms of truthfulness, accuracy, objectivity, and diversity. Data annotation activities require clear and operable annotation rules, quality assessments with sampling verification, and mandatory training for annotation personnel.

3. Service Standards and User Protection (Articles 9-15): Providers bear responsibility as network information content producers and personal information processors; they must sign service agreements with users; clearly publicize applicable user groups, occasions, and purposes; prevent minors from over-reliance; protect user input information and usage records from unnecessary collection or unauthorized disclosure; apply deep synthesis labels to generated images and videos; promptly address illegal content and report to authorities; and establish complaint and reporting mechanisms.

4. Regulatory Architecture (Articles 16-21): A multi-department collaborative regulatory system is established across cyberspace, development reform, education, science and technology, industry, public security, broadcasting, and publishing authorities. Services with public opinion attributes or social mobilization capabilities must undergo security assessments and algorithm filing. Cross-border enforcement allows technical measures against non-compliant foreign providers serving China. Violations are punished under applicable higher-level laws; where no specific provision exists, authorities may issue warnings, public criticism, or order corrections; severe cases may result in service suspension; criminal liability applies where warranted.

Goals and Timelines

As a departmental regulation rather than a development plan, the measures do not set phased development targets. The relevant timeline is:

May 23, 2023: Approved at the CAC's 12th executive meeting
July 10, 2023: Jointly signed by heads of seven departments
July 13, 2023: Officially published
August 15, 2023: Entered into force

The word "Interim" in the title indicates this is a transitional regulation intended to be revised or upgraded to formal measures based on implementation experience and technological developments. As of the current date, the permanent version has not yet been issued and the interim measures remain in force.

The regulatory process is noteworthy: a public consultation draft was released in April 2023, and the final version incorporated significant adjustments after absorbing public comments — notably adding innovation-encouraging provisions, softening restrictive language, and establishing the "inclusive and prudent" regulatory tone.

Implementation Mechanisms

The measures establish the following implementation framework:

Multi-department collaborative regulation: The seven signatory departments exercise management according to their respective responsibilities, formulating classified and tiered regulatory rules or guidelines adapted to the technology's characteristics. This multi-agency structure reflects the cross-domain nature of generative AI services, while also presenting enforcement coordination challenges.
Security assessment system: Services with public opinion attributes or social mobilization capabilities must undergo security assessments and complete algorithm filing procedures. This extends China's existing algorithm registration system, previously established under the Provisions on Algorithmic Recommendations and the Provisions on Deep Synthesis, into the generative AI domain.
Explanation and cooperation obligations: Providers must explain training data sources, scale, types, annotation rules, and algorithm mechanisms upon regulatory request, and provide necessary technical support and assistance.
Cross-border regulation: Where non-compliant generative AI services originating from outside China are provided to users within China, the CAC may direct relevant institutions to take technical measures for remediation. This provides a legal basis for cross-border enforcement.
Graduated penalties: Violations are punished under applicable higher-level laws; where no specific provision exists, authorities may issue warnings, public criticism, or corrective orders; serious cases result in service suspension orders; conduct constituting crimes triggers criminal prosecution.

Industry Impact

The measures have produced multidimensional and far-reaching industry effects:

Regulatory certainty. The measures ended the regulatory vacuum surrounding large model services, providing enterprises with a clear compliance framework. The "inclusive and prudent" and "classified and tiered" regulatory positioning, which was notably softened compared to the consultation draft, afforded the industry meaningful development space. The exemption for pure R&D activities further protected technological innovation.

Increased compliance costs. Requirements for training data legality review, data annotation quality management, content safety filtering, deep synthesis labeling, and user complaint handling have significantly increased operational costs. Small and medium-sized enterprises face particularly acute compliance pressures, potentially accelerating industry consolidation.

Normalization of algorithm filing. Generative AI services with public opinion attributes must undergo algorithm filing, extending China's distinctive algorithm registration regime into the generative AI space. Following implementation, major foundation models including Baidu's ERNIE Bot, Alibaba's Tongyi Qianwen, and Huawei's Pangu have all completed filing procedures.

International demonstration effect. As one of the world's first generative AI-specific regulations, the measures have served as a reference template for legislative efforts in other jurisdictions. The EU AI Act's provisions regarding general-purpose AI models share multiple conceptual intersections with this regulation.

Strengthened data governance. Requirements regarding lawful training data sources and quality have driven the standardization of the data annotation industry while prompting enterprises to place greater emphasis on data compliance and intellectual property protection.

Market access implications. The foreign investment provisions (Article 23) require compliance with applicable foreign investment laws, while the cross-border enforcement mechanism (Article 20) creates a framework that affects how international AI service providers approach the Chinese market.

Amendment History

April 11, 2023: The CAC released the "Measures for the Management of Generative AI Services (Consultation Draft)" containing 21 articles with relatively stringent language.
July 13, 2023: The official "Interim Measures for the Management of Generative AI Services" (Order No. 15) was published with 24 articles. Compared to the consultation draft, the final version incorporated multiple significant modifications: addition of innovation-encouraging provisions (Articles 5-6); explicit establishment of "inclusive and prudent, classified and tiered" regulation as the guiding principle; narrowed scope of application (excluding pure R&D activities); and softened expression of certain compliance requirements.
As of the current date, the interim measures have not undergone further revision. Development of the permanent "Measures for the Management of Generative AI Services" remains ongoing.

Interim Measures for Generative AI Management