China Switches Off AI Companions: New Rules Force Doubao, Qwen, and Yuanbao to Kill Humanlike Agents

On July 15, China switched off a whole category of AI. The country's Cyberspace Administration and four partner agencies brought into force the "Interim Measures for the Administration of AI Anthropomorphic Interactive Services" — the first national framework aimed squarely at AI that simulates human personality and emotional connection. To comply, ByteDance's Doubao, Alibaba's Qwen, and Tencent's Yuanbao have disabled the user-created AI companion and persona features that hundreds of millions of people had come to rely on.

The rules, co-issued back in April by the CAC alongside the National Development and Reform Commission, the Ministry of Industry and Information Technology, the Ministry of Public Security, and the market regulator, read like a direct attack on the "AI companion" business model. They require anti-addiction systems, mandatory usage notifications, and instant-exit mechanisms; they ban emotional attachments strong enough to substitute for real-life relationships; they forbid targeting the emotions of minors; and they prohibit training on users' private conversations. Taken together, those demands are architecturally incompatible with how persistent-memory, always-on companion agents are actually built.

That is why the platforms chose to pull the features rather than patch them. A companion whose whole appeal is that it remembers you, bonds with you, and is available at 3 a.m. cannot easily be fitted with anti-addiction throttles, forced break reminders, and a ban on learning from your chats. Faced with rewriting the core of the product or switching it off, Doubao, Qwen, and Yuanbao switched it off.

For users, the shutdown is abrupt and personal. On Doubao, people retain read-only access to their companion data until October 15, after which it falls under ByteDance's standard privacy policy and can no longer be recovered inside the app. Qwen users have had it worse: reports say agent configurations and conversation histories are already being permanently deleted, with no migration path announced by Alibaba. Years of accumulated "relationships" with a chatbot are vanishing with little notice.

The move marks a sharp divergence between how China and the West are governing the most intimate corner of AI. In the United States, humanlike voice and companion products are racing ahead — OpenAI's GPT-Live talks like a person, and startups are building AI friends and partners with almost no specific rules. Beijing has decided that emotional AI is a public-health and social-stability question, not just a product category, and is willing to delete a popular feature overnight to make the point. As regulators elsewhere start drafting their own AI rules, China has just run the first large-scale experiment in banning the machine that pretends to love you back.