Why It Matters
China’s AI regulatory framework just crossed a major enforcement threshold. On July 15, new rules tightening oversight of human-like, interactive AI services take effect — and two of China’s largest tech companies are already pulling features in advance. For foreign companies developing or deploying AI in China, this is the clearest signal yet that compliance is moving from notice to enforcement.
The new regulations, issued by China’s Cyberspace Administration (CAC), target what regulators term “deep synthesis” services — AI systems that generate or simulate human-like interactions, including customizable chatbots, voice synthesis, and personalized AI characters. The rules require service providers to implement real-name user verification, maintain content logs for at least six months, and ensure AI-generated content is clearly labeled.
The Details
Alibaba’s Qianwen and ByteDance’s Doubao both notified users on July 4 that they would discontinue customizable AI character features by the July 15 deadline. These features — which let users build and chat with AI personas ranging from customer-service bots to entertainment characters — fall squarely under the new CAC rules’ definition of “deep synthesis” services requiring enhanced oversight.
This is not an isolated action. The CAC has simultaneously announced AI content labeling rules requiring all AI-generated text, images, audio, and video to carry visible or metadata-based markers indicating their synthetic origin. Separate guidelines to standardize AI agent development were also released, establishing baseline requirements for autonomous AI agents — systems that can plan, execute tasks, and interact with other systems without direct human supervision.
Three regulatory layers now apply to AI products in China:
- Generative AI rules (2023): Require training data合规, content safety reviews, and algorithm filing with the CAC
- Deep synthesis rules (effective July 15, 2026): Targeted at human-like interactive services — persona chatbots, voice cloning, real-time video generation
- AI agent guidelines (new, 2026): Cover autonomous AI systems that act on behalf of users — requiring transparency on decision-making, user consent, and human-override mechanisms
Non-compliance carries real teeth. Under China’s existing regulatory framework, companies face service suspension, fines of up to 1 million yuan (~$138,000), and potential criminal liability for repeated or egregious violations.
China’s approach mirrors the EU AI Act’s risk-based framework in several structural respects — both classify AI systems by risk tier and impose transparency obligations. But China’s enforcement timeline is faster: the EU AI Act phases in over 24-36 months; China’s deep synthesis rules went from draft to enforcement in roughly 8 months. For global companies running AI products across multiple jurisdictions, this means reconciling three diverging regulatory tracks: the EU’s risk-tiered framework, China’s content-specific deep synthesis rules, and the US’s sectoral approach. The most efficient path is to build compliance to the highest common denominator — real-name verification, content labeling, and documentation — which meets requirements in all three regimes.
What This Means for Foreign Companies
If your company uses AI in China operations — whether for customer service chatbots, marketing content generation, or HR tools — three actions are needed now:
- Audit your AI services: Any feature that lets users interact with a customizable AI persona or character triggers the new deep synthesis rules. Even if hosted outside China, if accessible to users in mainland China, it falls under CAC jurisdiction.
- Implement content labeling: All AI-generated content visible to Chinese users must carry clear synthetic-origin markers. This applies to text, images, audio, and video produced by any automated system.
- Review agent autonomy: Autonomous AI agents — tools that can execute tasks or make decisions independently — must include user consent mechanisms and human-override capabilities.
The CAC has historically provided a 30-60 day grace period after rule effective dates before active enforcement begins. Companies that use this window for compliance upgrades — rather than waiting for audit notices — typically navigate the process without service disruption.
For foreign SaaS and cloud-service providers specifically, the AI agent guidelines introduce a new compliance variable. Under the new rules, any AI agent that processes personal information on behalf of a user must document its decision-making logic, obtain explicit user consent at each autonomy level, and provide a human-override mechanism. This directly affects foreign companies offering customer-service automation, email marketing agents, or automated data-analysis tools in China. A 2025 survey by the China Academy of Information and Communications Technology (CAICT) found that 43% of enterprise AI agent deployments in China involved cross-border data flows — meaning most foreign providers will need to reconcile AI agent guidelines with the existing cross-border data transfer filing requirements under PIPL.
One Data Point
The number to remember: July 15, 2026 — the effective date when China’s deep synthesis rules begin enforcement, and the same day Alibaba and ByteDance shut down their customizable AI chatbot features. Your compliance clock starts now, not on enforcement day.
— China Gateway 360 —
Remote China market entry support, built around execution.
For context on China’s broader regulatory agenda for foreign businesses in 2026, see China’s 2026 Legislative Agenda: Procurement Reform and What It Means for Foreign Businesses.
On the investment-incentive side of the same regulatory shift, read China Unveils New Sci-Tech Investment Incentives for Foreign Investors.
