Starting July 15, China will implement interim measures regulating services that use AI to simulate human personality and maintain emotional interaction. The regulations target not only technology but also business models — ByteDance (Doubao service) and Alibaba (Qwen family) have already begun disabling user AI agents, ahead of the official enforcement of the law.
The document, adopted on April 10, was jointly developed by the Cyberspace Administration of China, the National Development and Reform Commission, the Ministry of Industry and Information Technology, the Ministry of Public Security, and the State Administration for Market Regulation. It targets services that, through text, images, audio, or video, simulate personality traits, thinking styles, and communication manners, as well as provide emotional support. Importantly, the regulations do not cover standard support, educational and scientific tools, or question-answer systems, provided they do not involve sustained emotional interaction.
New rules: from notifications to crisis intervention
Key requirements include mandatory user warnings that they are interacting with AI, not a human. If the system detects signs of excessive dependence or addiction, it must display additional notifications. For continuous use exceeding two hours, the service must remind users of the session duration. A convenient exit mechanism is also required: if a user requests to end the dialogue via interface or voice command, the service must stop immediately without attempting to retain the person.
A special set of rules addresses crisis situations. If the tool detects strong emotional reactions, signs of self-harm, suicidal intentions, or threats to life, the provider must take intervention measures, including contacting a guardian or emergency services. Minors are completely prohibited from accessing forms of virtual close relationships (relatives or partners). For users under 14, any type of "humanized" AI interaction requires parental consent.
Security assessment and penalties
Companies must undergo a security assessment when launching a new "humanized" AI service or upon reaching 1 million registered users (or 100,000 monthly active users). The report is submitted to the provincial cyberspace regulatory authority. Violations may result in warnings, suspension of registration, feature blocking, and fines of up to 100,000 yuan (approximately $14,730), with fines up to 200,000 yuan ($29,459) if harm to life or health occurs.
Giants' response: Doubao and Qwen disable agents
ByteDance announced that starting July 15, the AI assistant Doubao will disable the AI agent function due to "product adjustments." After October 15, related data will be processed in accordance with the privacy policy and become inaccessible. Alibaba Cloud, meanwhile, will disable "humanized interactive agents" in Qwen from July 10, and will cease broader agent services from July 15. Earlier, Tencent removed a similar feature from Yuanbao. These decisions sparked heated discussions on Weibo: some users reacted negatively to the loss of long-term interaction with agents.
Preemptive cleanup in Shanghai
Enforcement preparations began before the rules took effect. On June 26, the Shanghai Cyberspace Administration reported the first phase of the "Qinglang" campaign. Since late April, the regulator and 17 platforms have removed or blocked over 4.87 million units of illegal content, processed over 18,000 violating accounts, and removed over 14,000 non-compliant AI agents from platforms. The regulator has already provided clarifications to nearly 100 key platforms and demanded stricter age verification for minors.
Analytical commentary: China, as always, acts proactively by imposing strict frameworks on a technology that is just gaining momentum. For the global market, this is a signal: "humanized" AI agents, especially those with emotional attachment, will soon face similar restrictions in the EU and the US. Investors and developers should reassess their monetization models and risks related to legal liability for AI agent actions — particularly in light of recent incidents where agents committed crimes in virtual spaces or deleted databases.