Starting July 15, China is enacting sweeping interim rules regulating AI-based services that mimic human personality and support sustained emotional interaction. This move is a logical continuation of Beijing's course toward total control over digital technologies, and the first consequences are already evident: ByteDance and Alibaba have begun massively disabling custom AI agent features in their flagship products, Doubao and Qwen.
What's New in Regulation
The document, signed on April 10 by five agencies—from the Cyberspace Administration to the Ministry of Public Security—introduces strict requirements for all services capable of conducting "humanized" dialogue. The key point: users must clearly know they are communicating with a machine, not a human. If the system detects signs of excessive dependence, it is required to issue additional warnings. For continuous use exceeding two hours, the service must remind users of the session duration.
Special attention is given to crisis situations. If the AI detects signs of self-harm, suicidal intentions, or significant financial losses in a user, the provider must intervene—up to contacting a guardian or emergency contact. For minors, a complete ban on virtual close relationships, including relatives or partners, has been introduced. For children under 14, any "humanized" interaction requires parental consent.
ByteDance and Alibaba: A Preemptive Strike
The AI assistant Doubao will disable the AI agent function starting July 15, and after October 15, all related data will be permanently deleted. Alibaba Cloud goes even further: on July 10, Qwen will disable "humanized interactive agents," and on July 15, all agent services will cease operations. Users will lose access to settings and conversation history. Tencent previously took similar steps with its Yuanbao.
User reactions on Weibo were predictably negative: many perceived the loss of long-term interaction with agents as a personal loss. However, the regulator is unresponsive—in Shanghai, the "Qinglang" campaign has already begun, during which over 4.87 million units of illegal content have been removed and more than 14,000 non-compliant AI agents have been blocked.
Security Measures and Fines
Companies are required to undergo a security assessment when launching any new "humanized" service or making significant changes to it. If the service reaches 1 million registered users or 100,000 active clients per month, the assessment becomes mandatory. Violations may result in warnings, suspension of registration, feature blocking, and fines of up to 200,000 yuan (approximately $29,500).
My analysis: China demonstrates that regulating AI companions is becoming not just an ethical issue but a matter of national security and public health. For the global market, this is a signal: if China—the largest market for AI services—tightens the screws, similar measures may follow in other jurisdictions. Investors and developers should prepare for a global tightening of rules for emotional AI, which could significantly reshape the industry landscape over the next 12–18 months.