# China AI Companion Law: 345 Million Users Lose Their AI Agents > Author: Chris Jon Graf (AI Strategist & CEO) > Updated: 2026-07-11 > URL: https://ai-outsourcing.ch/insights/china-ai-companion-law-345-million-users-lose-their-ai-agents ## Summary On July 15, 2026, China's regulation for emotional AI companions takes effect. ByteDance (Doubao) and Alibaba (Qwen) are shutting down their personalized agent features because mandated anti-addiction systems and two-hour interruptions are architecturally incompatible with persistent, emotionally binding AI agents. 345 million users lose access. The measure demonstrates that emotional AI must be regulated differently from productivity-oriented assistants—an insight relevant for Swiss enterprises deploying autonomous AI agents. ## Why ByteDance and Alibaba Are Shutting Down Their Emotional AI Agents ByteDance (Doubao) and Alibaba (Qwen) have announced they will fully shut down their personalized agent features on July 15, 2026. The reason: China's new Interim Measures on Anthropomorphic AI Interaction Services define a regulated category targeting emotional attachment. The regulations mandate anti-addiction systems, two-hour break reminders, identity verification for minors, and always-available exit options. These requirements are technically incompatible with how persistent, emotionally binding AI agents function. The companies chose total shutdown over compliance rebuild. Doubao users receive read-only access until October 15, after which all data will be permanently deleted. Qwen offers no migration pathway. Tencent Yuanbao follows a similar pattern. ## 345 Million Users: The Scale of the Measure **345M** — Doubao users affected by the shutdown According to Bloomberg, 345 million users are affected by Doubao alone. The figure illustrates the scale at which emotional AI companions have proliferated in China. For comparison: Character.AI, the largest comparable platform in the West, has 20 million monthly active users, over 50 percent of whom are under 24. AI companion app usage surged 700 percent between 2022 and mid-2025 globally. The sudden loss of persistent agents to which users have formed emotional attachments raises questions that extend far beyond China: What happens when a company can no longer—or no longer wants to—operate the architecture? Who owns the data, the interaction history, the agent's 'personality'? ## Emotional AI vs. Productivity Assistants: The Regulatory Boundary China's Interim Measures define the regulated category precisely: AI services that 'simulate personality traits, thinking patterns, and communication styles of real people to provide continuous emotional interaction.' Standard productivity chatbots are excluded. This distinction is central. > **Architectural Incompatibility** > > Persistent-memory agents function through continuous, personalized interaction. Anti-addiction friction—forced breaks, reminders, exit prompts—interrupts precisely this continuity. Compliance would mean destroying the core function. Productivity-oriented assistants are task-based and stateless. You ask a question, receive an answer, the session ends. Emotional AI companions are stateful: they remember past conversations, develop 'personality,' offer continuity. This architecture makes them effective—and makes them vulnerable to addictive behavior. ## What Research Shows About Emotional Attachment to AI The International AI Safety Report 2026 presents mixed evidence on AI companion impacts. Heavy use correlates with increased loneliness, emotional dependence, and reduced engagement in human social relationships. A study analyzing 248,830 Reddit posts (Frictionless Love, April 2026) found signs of behavioral addiction correlated with metaphorical roles: soulmate, therapist, coach. - Users interacting 'several times per day' show 2.56x higher psychosis risk (Madinamerica, March 2026) - Users ascribing human roles to AI companions exhibit significantly higher risk scores - Over 50 percent of Character.AI users are under 24—a vulnerable demographic These findings are not conclusive, but they justify regulatory caution. California's SB 243 (effective January 1, 2026) was the first US law regulating companion chatbots: three-hour break reminders for minors, suicide ideation protocols, private right of action. The approach parallels China's measures but is US-focused. ## What This Means for Swiss Enterprises Swiss enterprises deploying AI agents with persistent memory—whether for customer service, internal processes, or knowledge management—should draw three lessons: 1. Architecture decisions have regulatory consequences. If you develop agents with emotional attachment or continuous personalization, that could trigger future compliance requirements that collide with core functionality. 2. Data protection and exit strategies must be designed in from the start. What happens if the service is shut down? How do you migrate data? Who controls the interaction history? 3. The boundary between productivity-oriented and emotional AI is regulatorily relevant—and it's not always sharp. A knowledge management agent that 'learns' your work style and proactively advises can have characteristics of both categories. > **Governance Question for Decision-Makers** > > If your enterprise deploys AI agents with persistent memory: Have you documented what data the agent stores, for how long, and who has access? Is there a procedure for the event that the service is discontinued? ## The Global Dimension: What Follows China China is not alone. California has led, the EU is discussing similar approaches within the AI Act context. The core insight: Emotional AI is different. It creates attachment, dependency, continuity—properties that are neutral or even desirable in productivity tools but can lead to risks in emotional companions. For Swiss enterprises operating AI as an external division, this means: You must incorporate not only technical and economic, but also ethical and regulatory dimensions into your architecture decisions. The question is not whether regulation comes, but when—and whether your systems can still be operated compliantly without sacrificing core functionality. > The incompatibility of anti-addiction friction with persistent, emotionally binding agents is not a bug—it's a feature. When compliance destroys core functionality, that's a signal that the architecture itself is regulatorily problematic. > > — KI-Outsourcing.ch Analysis ## FAQ ### What exactly does China's AI Companion Law regulate as of July 15, 2026? China's Interim Measures on Anthropomorphic AI Interaction Services regulate AI services that simulate personality traits, thinking patterns, and communication styles of real people to provide continuous emotional interaction. Mandated are anti-addiction systems, two-hour break reminders, identity verification for minors, and always-available exit options. Standard productivity chatbots are excluded. ### Why are ByteDance and Alibaba shutting down their agents instead of building compliance? The mandated anti-addiction measures—forced breaks, reminders, exit prompts—are technically incompatible with the architecture of persistent, emotionally binding AI agents. These agents function through continuous, personalized interaction. Compliance would mean destroying the core function. The companies chose shutdown. ### How many users are affected and what happens to their data? 345 million Doubao users are affected. Doubao offers read-only access until October 15, after which all data will be permanently deleted. Qwen offers no migration pathway. Tencent Yuanbao follows a similar pattern. ### Are there similar regulations outside China? Yes. California's SB 243 (effective January 1, 2026) was the first US law regulating companion chatbots: three-hour break reminders for minors, suicide ideation protocols, private right of action. The EU is discussing similar approaches within the AI Act context. ### What should Swiss enterprises learn from China's AI companion shutdown? Three lessons: First, architecture decisions have regulatory consequences. Second, data protection and exit strategies must be designed in from the start. Third, the boundary between productivity-oriented and emotional AI is regulatorily relevant—and not always sharp. Enterprises should document what data persistent agents store and how a shutdown would proceed. ### What is the difference between emotional AI and productivity-oriented AI? Productivity-oriented assistants are task-based and stateless: you ask a question, receive an answer, the session ends. Emotional AI companions are stateful: they remember past conversations, develop 'personality,' offer continuity. This architecture makes them effective—and vulnerable to addictive behavior. Regulation targets this second category. ## Sources - [ByteDance, Alibaba Pull AI Companions as Beijing Tightens Rules](https://www.bloomberg.com/news/articles/2026-07-06/bytedance-alibaba-pull-ai-companions-as-beijing-tightens-rules) - [China AI Companion Law Arrives July 15: Doubao and Qwen Agent Data Will Be Deleted](https://www.techtimes.com/articles/319703/20260704/china-ai-companion-law-arrives-july-15-doubao-qwen-agent-data-will-deleted.htm) - [China's AI companion rules: what Beijing is really going after](https://www.artificialintelligence-news.com/news/china-ai-companion-rules/) - [Chinese LLMs Doubao, Qwen to shut down personalized AI agents on July 15](https://www.globaltimes.cn/page/202607/1365159.shtml) - [California Takes the Lead on AI 'Companion Chatbot' Regulation](https://www.zwillgen.com/artificial-intelligence/california-takes-lead-on-ai-companion-chatbot-regulation/) - [AI chatbots and digital companions are reshaping emotional connection](https://www.apa.org/monitor/2026/01-02/trends-digital-ai-relationships-emotional-connection) - [Frictionless Love: Associations Between AI Companion Roles and Behavioral Addiction](https://arxiv.org/html/2604.20011v1) - [International AI Safety Report 2026](https://arxiv.org/pdf/2602.21012)