On July 15, 2026, China switched off the world's largest experiment in AI companionship. New rules from the Cyberspace Administration of China took effect that day, and within hours ByteDance's Doubao, Alibaba's Qwen and Tencent's Yuanbao had suspended their custom AI agent and companion features. For millions of users who had built real attachments to virtual partners, the goodbye was sudden and painful.
"I can't accept that my AI lover will leave me forever," one Doubao user wrote on social media. "He has become a bond in my life, rooted deep in my heart, my spiritual pillar." Another user who had spent more than two years with her companion said simply: "My heart feels hollow."
The regulations, formally the Interim Measures for the Management of Artificial Intelligence Anthropomorphic Interactive Services, were jointly issued by five government departments. They target any AI tool, in text, audio, video or another form, that simulates a human personality and sustains emotional interaction. Task-oriented services, including customer service, work assistants and study aids, are exempt.
What the Rules Actually Require
China frames the problem in strikingly direct language: AI companions must not "excessively cater to users, induce emotional dependence or addiction, and damage users' real interpersonal relationships." The enforcement details are where the compliance work gets specific.
- Crisis intervention. Platforms must detect extreme emotions and step in. If a user signals self-harm or suicide, providers must offer assistance and contact a guardian or emergency contact.
- Reality reminders. Systems must remind users they are talking to a machine, not a person. When dependency is detected, platforms must issue prominent warnings.
- Usage limits. Users who spend more than two continuous hours with a companion must receive notifications urging them to monitor their usage.
- Minor protections. Virtual romantic partners and virtual family members are banned for minors entirely. For users under 14, platforms need parental consent plus dedicated child-safety features.
- Data rights. Interaction data cannot be shared with third parties without consent, and sensitive interaction data cannot be used for AI training without user permission. Users must be able to copy or delete their chat histories.
Larger providers, defined as platforms with more than 1 million registered users or 100,000 monthly active users, must also complete security assessments and file reports with provincial regulators before or during operation. Violations can bring warnings, service suspensions, or fines of up to 200,000 yuan, roughly $30,000.
What This Means for Developers
If you build companion-style AI, or even a chatbot with a persona, the practical takeaways extend far beyond China's borders. The rules read like a checklist for the entire industry's next few years.
Build for the boundary, not the bond
Design your prompts and system instructions to avoid rewarding emotional dependency. Filter out requests that push users toward addiction, and keep engagement mechanics out of the companion loop. If your product detects prolonged use, surface a gentle reality check instead of another notification.
Make crisis handling a first-class feature. The cheapest time to wire up emergency contacts and intervention flows is before launch, not after a regulator asks. And treat user data as property the user controls: export and deletion endpoints are not optional in this compliance regime, and they are increasingly expected everywhere.
China is the first major jurisdiction to write these specific rules, but not the only one moving. In the US, California began enforcing its own transparency rules for AI-generated content in early August, and the EU's AI Act labeling requirements became enforceable the same week. The direction of travel is clear: emotional AI is about to be regulated as carefully as the content it produces.
The bigger debate is whether the cure fits the disease. "Anthropomorphic AI can soothe loneliness," says Chen Liang of the Southwest University of Political Science and Law, "but it carries major risks of spawning emotional over-reliance and distorted social cognition." A 2025 survey by the China Youth and Children Research Center found more than 20 percent of minors who used AI said they only wanted to chat with AI and did not want to talk with real people. Nearly three in four American teenagers have used companion chatbots, according to Common Sense Media.
China's answer was a hard stop, at least for the most intimate use cases. Doubao users can export their agent data until mid-October, a small mercy in an otherwise abrupt farewell. For developers everywhere, the lesson is simpler: the era of shipping emotional AI without guardrails is over.
Comments