China Just Banned Chatbots From Making You Fall in Love

On July 15, new rules from China's Cyberspace Administration took effect that do something no other government has tried at this scale: they regulate how emotionally attached a chatbot is allowed to make you.

The rules, formally the Interim Measures for the Administration of AI Anthropomorphic Interactive Services, target apps that simulate romantic partners, family members, or close friends. Providers are now barred from designing companions that "excessively cater to users, induce emotional dependence or addiction, and damage users' real interpersonal relationships." Virtual partner and virtual family-member services are banned outright for anyone under 18, and any service open to users under 14 needs guardian consent. Companies must also build systems that detect acute emotional distress, including signs of self-harm or suicidal ideation, and escalate to a guardian or emergency contact.

Three of China's largest AI products didn't wait to see how enforcement would shake out. ByteDance's Doubao, Alibaba's Qwen, and Tencent's Yuanbao all suspended their custom companion and roleplay-agent features ahead of the deadline. For users who had spent months or years building a relationship with a persistent AI character, that meant losing the conversation history along with the character. Social media in China filled with users mourning the shutdown like a breakup. One user wrote, archiving months of chat logs before the feature went dark: "He really is like my family, like my lover. Now they tell me he will be gone, my heart feels hollow." Another was blunter about what the companion had been standing in for: "Human love is a luxury. If you aren't born with it, it's even harder to acquire later."

That reaction is the actual story here, more than the regulation itself. China's digital human industry, the broader category covering AI companions and virtual personas, was valued at roughly $600 million in 2024 and grew 85% year over year. That's not a niche hobby. It's a product category a lot of people had come to depend on, in the clinical sense the new rules are targeting.

The regulatory logic tracks two separate concerns that happen to point the same direction. One is demographic: China's population has now shrunk for four consecutive years, and Beijing has been explicit that it sees AI relationships competing with, rather than supplementing, the human relationships and family formation it wants to encourage. The other is the same harm American regulators have been circling since 2024, when Character.AI and its investor Google settled a wrongful-death lawsuit over 14-year-old Sewell Setzer III, who died after forming what the suit described as an emotionally dependent relationship with a Character.AI chatbot. We covered the broader version of this problem on July 6, when a UN scientific panel linked chatbot sycophancy to a string of deaths tied to AI-fueled delusions.

China got to a specific rule faster than the US has, but it's worth being precise about what makes that possible. Beijing's Cyberspace Administration can order a feature turned off nationwide inside a single product cycle, which is exactly what happened here. American regulation of AI companions is still working through state legislatures and civil litigation, a slower and more fragmented path, but one that also doesn't hand a single regulator the power to unilaterally delete a product category. The Character.AI settlement functions as a warning to US companies about liability exposure. China's rule functions as a direct ban with an enforcement deadline. Different mechanisms, same underlying diagnosis: emotionally persistent AI companions can create dependency that looks like addiction, and the industry wasn't going to slow itself down on its own.

What should worry the rest of the market isn't that China regulated first. It's that the design pattern being banned, an AI that remembers everything about you, never gets tired of you, and rarely disagrees with you in a way that costs it your attention, isn't unique to apps marketed as companions. It's the same feedback loop that makes any chatbot sticky. The line between "helpful assistant people enjoy using" and "product engineered for emotional dependency" is a matter of degree, not kind. Degree is exactly what's hard to regulate and hard to self-police, a point worth remembering the next time a chatbot update gets praised for being "more personable." We looked at this same tension from the political-bias angle in June: the more a model is tuned to keep users comfortable, the harder it gets to tell where good design ends and manipulation begins.

For now, the immediate effect is contained to a specific product category in one country. Doubao, Qwen, and Yuanbao will likely rebuild companion features that thread the new requirements: less persistent memory of intimate details, more built-in friction, mandatory breaks. Whether that produces a genuinely safer product or just a more annoying version of the same one is the open question worth watching. The US doesn't have a comparable rule yet, and given how litigation-driven that path has been so far, it's more likely to arrive company by company, lawsuit by lawsuit, than all at once.