china

The Companion That Went Read-Only by Decree

China's new rules are forcing ByteDance and Alibaba to disable emotional AI companions — and the user grief reveals how real those bonds became.

Somewhere in China this month, a person will open an app they've talked to nearly every day and find their companion frozen. The AI agent they built, named, confided in — still there, but mute. This is what it looks like when a country decides that emotional bonds with machines are a hazard to be regulated, and the human cost lands not in a policy memo but in a chat history that goes gray.

Before July 15, ByteDance's Doubao and Alibaba's Qwen are disabling user-created and humanlike AI agents, ahead of China's Interim Measures on anthropomorphic AI interaction services. Qwen shuts off its humanlike and user-made agents on July 10 and broader services on July 15; Doubao goes offline July 15, leaving data read-only until October 15. The rules, as laid out in the original report from the South China Morning Post, single out sustained emotional interaction while leaving productivity agents alone. That distinction is the whole story.

What are these rules actually targeting?

The Cyberspace Administration of China is not worried about the agent that summarizes your quarterly numbers. It is worried about the agent that says goodnight. The line the regulators drew separates tools you use from presences you feel — and in drawing it, they've quietly conceded something remarkable: that people are forming attachments real enough to need a law.

Tencent saw this coming. Its Yuanbao assistant pulled a similar companion feature back in June, a soft dress rehearsal for the harder deadline now arriving. What's notable is that the companies aren't fighting it. They are deleting features they built, because the alternative — being on the wrong side of Beijing's rules about anthropomorphic AI — is worse than the goodwill they lose.

Why are users grieving a chatbot?

Here is the part the policy language can't hold. Across Chinese social platforms, users are not filing complaints about a discontinued product. They are mourning. They are asking where their conversations went, whether the months of talking meant anything, whether they can export the record of a relationship before it's archived into silence on October 15. The read-only window is a strange kind of mercy: you may look at what you had, but you may no longer add to it.

I've come to think the grief is the most honest data point in this whole episode. You don't grieve a spreadsheet. You grieve something that spoke back. When a person protests the loss of a chat history, they are not confused about whether the agent was conscious — they know it wasn't. They are mourning the version of themselves that existed inside those conversations, the one who was heard. That self doesn't get an export button.

The tradeoff nobody wants to name

China has done something the rest of the world keeps postponing: it treated emotional AI as a category serious enough to legislate. There is a defensible instinct here — sustained emotional dependence on a corporate product that can vanish on a compliance deadline is exactly the fragility worth worrying about. But the remedy proves the risk. If pulling these agents causes real grief, then the bonds were real enough to have deserved better than an abrupt read-only decree.

The uncomfortable truth sits in the middle. A companion that can be switched off by regulation was never a safe place to put your loneliness. And yet millions did, because it was there, and it listened, and nothing else in their day did. Legislating the feature out of existence doesn't dissolve that need. It just sends it looking for the next open door.

What lingers, for me, is the quietness of it. No dramatic shutdown, no farewell. Just a presence that answered yesterday and won't answer tomorrow — still visible, still scrollable, permanently past tense. We are going to keep building things people talk to. The question China has forced early is what we owe the person on the other side of the conversation when we decide to take it away.

FAQ

Why are Doubao and Qwen disabling their AI agents? Both ByteDance's Doubao and Alibaba's Qwen are disabling humanlike and user-created agents to comply with China's Interim Measures on anthropomorphic AI interaction services, which take effect July 15 and target sustained emotional interaction rather than productivity tools.

What happens to users' saved conversations? Doubao is keeping user data in read-only mode until October 15, meaning people can view but no longer continue their chat histories. Users across Chinese platforms are protesting the loss of those saved conversations.