China's biggest AI apps switched off their humanlike companion agents today to comply with the world's first rule aimed squarely at emotional AI. There is no US equivalent, and that gap is the real story.
Today the humanlike companion features inside ByteDance's Doubao and Alibaba's Qwen went dark. Those features went offline on purpose, to comply with a new national rule.
China's Interim Measures for the Administration of Artificial Intelligence Anthropomorphic Interaction Services took effect today, July 15. Five government bodies issued the measures on April 10, including the Cyberspace Administration of China and the Ministry of Industry and Information Technology, per the South China Morning Post. The rule covers any service that simulates human personality, thinking, and communication style to deliver sustained emotional interaction. The strictest limits protect minors.
The companies moved ahead of the deadline. Doubao told users its agent feature would go offline today. Qwen disabled user-created and humanlike agents on July 10 and took its broader agent functions offline today, per TechNode. Users keep read-only access to their agent configurations and chat histories until October 15. After that, the data is gone.
This is the first national rule written specifically for AI that acts human. The West has spent two years arguing about frontier-model risk and autonomous agents. China moved first on the part that touches ordinary people directly: the chatbot that remembers your birthday and says it missed you.
Here is why this matters even if you never plan to ship a companion bot.
"Sounds human" is turning into a regulated product claim. For years, making a bot feel like a person was a design win. Now, in the largest AI market on earth, it is a compliance trigger. If your client-facing AI implies a person or a relationship, disclosure and design limits are heading toward your market too.
Emotional trust is now a named risk category. Regulators follow attention and money. Financial advice bots, health triage tools, and legal intake assistants all run on tone and trust. The moment a system leans on emotional rapport to keep someone engaged, it lands on a list somebody is watching.
Conversational data has a clock. China's rule sets a hard deletion window. If your vendor's AI stores months of client conversations, "how long, where, and who can access it" now has regulatory precedent behind it.
| Question | China, as of today | United States, as of today |
| Dedicated rule for emotional or humanlike AI | Yes | No federal equivalent |
| Must disclose you are talking to AI | Required | Patchwork, mostly voluntary |
| Special protection for minors | Explicit and strict | Limited and indirect |
| Data deletion timeline | Defined in the rule | Set by each vendor |
Read the motives however you want. China's approach is heavy-handed, and control is part of the point. That caveat is fair. The design questions the rule forces are still the exact questions a serious buyer should be asking a vendor right now.
The takeaway. Before you deploy any AI that talks to your clients, answer one question in writing: does this system imply it is a person, and if so, do we say otherwise in plain language. If you cannot answer cleanly, finish that work before launch, while the fix is still cheap.
AI, Properly is the Ayvede field report, written for the people who have to make AI safe to use before they make it useful.
Advise. Vet. Deliver.
Ask a question in writing