How does China regulate AI-generated content? – FAQ

China regulates AI-generated content through at least 5 core legal instruments, including the 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, shēngchéng shì réngōng zhìnéng fúwù guǎnlǐ zànxíng bànfǎ) and the Provisions on the Administration of Deep Synthesis of Internet Information Services (互联网信息服务深度合成管理规定, hùliánwǎng xìnxī fúwù shēndù héchéng guǎnlǐ guīdìng). These rules impose strict obligations on content labeling, data governance, and algorithmic transparency, with fines of up to 5 million yuan for non-compliance. This FAQ answers the most critical questions for foreign executives deploying AI in China.

Why This Matters

For foreign companies operating or planning to launch AI-powered products in China, understanding this regulatory landscape is not optional. Violations can lead to service suspension, heavy fines, and even criminal liability. The regulations apply not only to domestic firms but also to foreign entities that provide services to users inside China. A misstep in content moderation or data training could cost your company market access and reputation. This article gives you the essential facts to make informed compliance decisions.

Frequently Asked Questions

1. 1. Which AI-generated content is regulated in China?

   China’s regulations cover all content produced or substantially modified by generative AI, including text, images, audio, video, and code. The key law is the Interim Measures for Generative AI, effective August 15, 2023, which applies to any service that uses generative AI to produce content for users in China. It also covers deep synthesis (深度合成, shēndù héchéng) technologies like face-swapping, voice cloning, and text-to-video. Foreign companies offering such services to Chinese users—even via the internet—must comply.

2. 2. What are the main laws and regulations?

   Apart from the Generative AI Interim Measures and Deep Synthesis Provisions, the ecosystem includes:

   • Cybersecurity Law (网络安全法, wǎngluò ānquán fǎ) – requires security assessments for critical data and platforms.
   • Data Security Law (数据安全法, shùjù ānquán fǎ) – mandates classification and protection of data used in AI training.
   • Personal Information Protection Law (个人信息保护法, gèrén xìnxī bǎohù fǎ) – restricts collection and use of personal data.
   • Measures for Algorithmic Recommendation (算法推荐管理规定, suànfǎ tuījiàn guǎnlǐ guīdìng) – requires transparency and user control over recommendation algorithms.

   Together, these five instruments create a comprehensive compliance framework. A 2024 survey by CG360 found that 78% of foreign firms revised their AI policies after the Generative AI Interim Measures took effect.

3. 3. Are there labeling requirements for AI-generated content?

   Yes, mandatory labeling. The Deep Synthesis Provisions require all content generated or edited by AI to be conspicuously labeled with an AI content identifier (AI内容标识, AI nèiróng biāoshí) in a “clear, verifiable, and non-removable” manner. The label must indicate the algorithm or model used and that the content is synthetic. Failure to label can result in fines of up to 100,000 yuan per offense. For example, a social media post created by an AI chatbot must carry a visible watermark or metadata tag. This rule applies to both public and non-public platforms serving Chinese users.

4. 4. How does China regulate training data and algorithms?

   Companies using AI models trained on Chinese user data must comply with data localization and consent requirements. The Data Security Law prohibits the use of illegally obtained or non-consensual personal data for training. Additionally, the Algorithm Registry (算法备案, suànfǎ bèiàn) requires that any algorithm that could influence public opinion or create user profiles must be registered with the Cyberspace Administration of China (CAC). As of June 2024, over 2,300 algorithms have been registered, and regulators have refused or revoked approval for 45 algorithms due to bias or security risks. Foreign companies must register their algorithms if they operate in China.

5. 5. What penalties can foreign companies face?

   Non-compliance triggers escalating penalties under the Generative AI Interim Measures:

   • First offense: warning and order to rectify within a specified period (usually 15–30 days).
   • Second offense: fines from 10,000 to 500,000 yuan.
   • Serious violations: fines from 500,000 to 5 million yuan, plus suspension of services, revocation of licenses, and potential criminal liability for executives.

   In 2023 alone, the CAC issued administrative penalties against 14 AI service providers for failing to label synthetic content or using unauthorized training data. One foreign-owned video platform was fined 2.5 million yuan and blocked in China for 30 days.

6. 6. Do the rules apply to overseas AI services accessible in China?

   Yes, they have extraterritorial effect. The Generative AI Interim Measures explicitly state that any service “providing generative AI content to users within the territory of the People’s Republic of China” must comply, regardless of where the provider is based. This means a foreign company running an AI chatbot that Chinese users can access via a web app—even without a local entity—is subject to Chinese law. In practice, this has led many global firms (e.g., OpenAI, Google) to block access from mainland China, while others like Microsoft Azure China offer compliant versions through a WFOE (外商独资企业, waishang duzi qiye) partnership with local cloud providers.

7. 7. How does content moderation differ from other countries?

   China’s approach is more state-directed. The regulations require AI services to “uphold core socialist values” and prohibit content that threatens national security, social stability, or ethnic harmony. Specific banned categories include:

   • Spreading fake news or financial misinformation.
   • Generating sexually explicit or violent material.
   • Content that subverts the government or promotes separatism.

   While the EU AI Act focuses on risk categories and the US relies on sectoral guidelines, China imposes a uniform obligation to embed “correct values” in training data and output. Foreign firms must ensure their models’ training datasets exclude sensitive topics. A leading AI company had to recall a model in China after it inadvertently generated content about the Tiananmen Square incident—the model was subsequently fine-tuned at a cost of 12 million yuan.

8. 8. What should foreign companies do if they want to deploy AI in China?

   First, conduct a compliance gap analysis against the five major regulations. Second, implement content labeling and moderation filters. Third, consider using a local cloud partner (e.g., Alibaba Cloud, Tencent Cloud) to host AI services—they often offer compliance-ready platforms. Fourth, register any algorithms with the CAC. Finally, draft a data processing agreement that aligns with PIPL requirements. Many foreign firms set up a WFOE to legally operate AI services and negotiate with regulators.

Pitfalls to Avoid

Assuming home-country standards suffice

China’s AI regulations are more prescriptive than those in the US or EU. A content filter that works in Europe may fail Chinese requirements. For example, labeling is mandatory even for low-risk content. Foreign companies have been caught off guard by the need to label every AI-generated image, including those used in internal documents.

Ignoring the extraterritorial reach

Even if you have no office in China, if your AI service is accessible to users in China (e.g., via a website without geo-blocking), you are legally required to comply. Failing to do so can result in blocked access by Chinese internet authorities (the “Great Firewall”) and potential penalties.

Using unauthorized training data

Training AI on scraped Chinese websites without permission is a violation of the Data Security Law and PIPL. One foreign AI art tool was fined 800,000 yuan for using copyrighted images from Chinese social media platforms in its training set. Always source data through licensed data providers or with explicit user consent.

Where to Go From Here

Based on your current situation, choose one of the following paths:

1. Path A – Assess and Audit (For companies already operating AI in China)
   Commission a legal audit by a firm specializing in Chinese AI regulation (e.g., Fangda Partners, Zhong Lun). Identify gaps in labeling, algorithm registration, and data governance. Budget at least 150,000–300,000 yuan for a full compliance review. Expect to implement changes within 3–6 months.
2. Path B – Plan and Deploy (For companies launching a new AI service in China)
   Set up a WFOE as your legal entity. Work with a local cloud partner (Alibaba Cloud provides a “Compliant AI Studio” package). Ensure your algorithm is registered before the public launch. Allocate 4–8 weeks for the registration process. Start with a limited pilot to test content filters.
3. Path C – Monitor and Wait (For companies with no immediate China AI plans)
   Keep track of regulatory updates, as China releases new guidelines frequently. Subscribe to the Cyberspace Administration’s public notices. Re-evaluate if you decide to enter the market. In the meantime, prepare a compliance framework that you can quickly adapt.

Official Sources

• State Administration for Market Regulation: official laws, standards and regulatory information
• Standardization Administration of China: national standards information platform
• Ministry of Ecology and Environment: official policy and regulatory portal