China began enforcing its AI content labeling regulations in late July 2026, requiring all AI-generated content — including text, images, audio, and video — to carry visible disclosures. For foreign technology companies operating AI-powered products or services in China, the new rules create immediate compliance obligations that cut across product development, marketing, and customer communication.
The Cyberspace Administration of China (CAC) issued the rules under the broader framework of the 2023 Interim Measures for the Management of Generative AI Services. The enforcement push comes alongside a related development: Beijing-based startup Moonshot AI open-sourced its Kimi K3 model on July 28, triggering renewed debate over AI safety labeling and intellectual property disclosure in cross-border contexts.
What the Rules Require
The regulations apply to any organization that “generates, publishes, or distributes AI-generated content accessible to users in China” — a definition that captures most foreign AI companies with Chinese users, including SaaS platforms, content tools, enterprise AI assistants, and marketing automation software.
| Content Type | Labeling Requirement | Where to Display |
|---|---|---|
| AI-generated text | Visible watermark or disclaimer | Beginning or end of content |
| AI-generated images | Digital watermark (visible or invisible) | Bottom-right corner minimum |
| AI-generated audio | Verbal disclosure at start | First 5 seconds of playback |
| AI-generated video | Text overlay throughout | Persistent corner label |
| AI-assisted edits | Disclosure on distribution | Metadata or platform tag |
Foreign Companies Most Affected
The rules are not new — they were first published in draft form in 2024. But the enforcement environment has shifted. CAC inspections in June and July 2026 resulted in formal warnings to at least 14 companies, including three foreign-invested enterprises, for non-compliance. The affected foreign firms included one European enterprise AI platform, one U.S.-based design tool provider, and one Japanese customer-service chatbot operator — illustrating the broad scope of enforcement.
- SaaS and enterprise AI platforms: Foreign companies offering AI-powered writing assistants, code generation tools, or design automation must add content labels to all output generated through China-accessible instances. This affects Copilot-style tools, Canva-type platforms, and automated copywriting services.
- Marketing automation providers: Any foreign tech company serving Chinese clients with AI-generated marketing copy, social media content, or ad creative must ensure labels appear before distribution. Third-party liability applies — the platform, not just the end user, bears compliance responsibility.
- Customer service chatbots: Foreign companies using AI chatbots for Chinese customer support — whether for e-commerce, travel, or financial services — must ensure the bot identifies itself as AI when initiating conversations. The verbal disclosure requirement for audio applies to voice-based customer service systems.
- Content creation tools: Video editing platforms, image generators, and music composition tools accessible in China must embed labeling at the generation point, not at the point of sharing. This means the labeling logic must exist within the tool itself.
Enforcement and Penalties
According to CAC guidance published alongside the enforcement push, penalties for non-compliance follow a graduated scale: written warnings and correction orders for first-time violations (observed in July 2026 cases), fines of up to ¥100,000 (US$13,800) for failure to correct within 30 days, and potential service suspension for repeated or willful violations. Foreign companies should note that the CAC has the authority to block non-compliant services from Chinese app stores and cloud platforms.
The enforcement push coincides with broader AI governance developments. Moonshot AI’s open-source release of Kimi K3 on July 28 — a model that some analysts say rivals GPT-4 in Chinese-language tasks — has drawn scrutiny over whether open-weight distributions fall under the labeling rules. The CAC has not yet issued specific guidance on open-source model labeling, but foreign companies distributing or fine-tuning open-source models in China should monitor this closely.
What Foreign Tech Companies Should Do
- Audit all AI-powered features accessible to users in China — including features used by Chinese subsidiaries of multinational corporations
- Implement labeling logic at the point of generation (text watermarks, image markers, audio disclosures, video overlays)
- Update terms of service and privacy policies to reflect AI content labeling obligations
- Create internal compliance documentation showing labeling implementation for potential CAC inspection
- For voice-based customer service: ensure AI disclosure within the first 5 seconds of any automated call
- Monitor CAC guidance on open-source AI model labeling — the regulatory environment around model weights remains in flux
One Data Point
The number to remember: 14 companies warned in 60 days. CAC enforcement of AI content labeling is not theoretical — it’s happening now. If your foreign tech company has Chinese users and generates AI content, the compliance clock is running.
Where to Go From Here
Based on what you just read:
- Ready to act? Read [guide: china-ai-compliance-checklist-2026]
- Still comparing? See [comparison: cac-ai-rules-vs-eu-ai-act-2026]
- Need numbers? Try [tool: ai-labeling-compliance-audit-tool]
— China Gateway 360 —
Remote China market entry support, built around execution.
