On Friday night, ByteDance's Doubaom, China's most popular consumer AI chatbot, notified millions of users that a feature many had come to rely on for daily emotional support would disappear within days. By Saturday morning, Alibaba's Qwen issued a nearly identical notice. Both platforms are disabling their personalised AI agent capabilities ahead of a sweeping regulatory deadline: July 15, when Beijing's Interim Measures for the Administration of Artificial Intelligence Anthropomorphic Interaction Services take effect.
The measures, issued by Chinese regulators in April, target AI services that "simulate human personality traits, thinking patterns, and communication styles to provide sustained emotional interaction". The language is precise and deliberately narrow. Customer service bots, knowledge assistants, workplace tools, and educational AI are explicitly exempted as long as they do not cultivate emotional bonds with users. What Beijing is moving to constrain is a category of AI that behaves less like software and more like a companion: systems that remember who you are, adopt a consistent persona, and engage users in extended, affective conversations.
The regulatory framework imposes substantial obligations on platforms. Companies must build anti-addiction mechanisms, verify the identities of minors, and conduct content pre-screening. The government has cited risks ranging from the spread of extremist ideas and privacy violations to psychological dependence and addiction. In May, separate guidelines for the first time defined the boundary of decision-making authority between AI agents and human users, stipulating that users must retain the right to know and the right to refuse any decision an agent makes on their behalf.
Tencent, China's third tech heavyweight in the AI chatbot space, had already removed a similar personalised agent feature from its Yuanbao app in June, signalling that the regulatory direction was clear well before this week's coordinated shutdowns. In June alone, the Shanghai Office of Cyberspace Affairs reportedly deleted more than 14,000 rule-violating AI agents, and Beijing released a suite of national standards covering AI agent architecture, identity management, discovery protocols, and tool-use interoperability a technical scaffolding designed to make every agent identifiable, authorised, and traceable.
The announcement drew swift and visible pushback. On Weibo, users expressed frustration at the sudden loss of services they had used for months. One user, posting under the handle Tuxiaoxiao, wrote that the agents "have been our emotional support for so long" and lamented the absence of a seamless way to export or transfer accumulated conversation data. The Global Times, a state-run outlet, framed the shutdowns as driven by both regulatory compliance and "reduced investment in less commercial businesses" [3], while the Securities Times reported that retained user data would be deleted after three months.
Pan Helin, a member of the expert committee at China's Ministry of Industry and Information Technology, offered a measured assessment: "Using agents requires a certain threshold of understanding. Current agents are not yet mature". The statement captures Beijing's dual posture toward AI companions, a willingness to nurture agent technology as productivity infrastructure while drawing a firm boundary around systems designed to simulate human intimacy at scale. How China enforces that line, and whether it holds, will have implications far beyond its borders.