China’s two mainstream consumer-facing artificial intelligence applications—ByteDance’s Doubao and Alibaba Group’s Tongyi Qianwen (Qwen)—are preparing to offline customizable anthropomorphic agent functions in preparation for the Interim Measures for the Management of Artificial Intelligence Anthropomorphic Interactive Services, which will officially take effect on July 15.
Doubao issued a notice to users on Friday night saying that its proxy function would be offline starting July 15, citing "product function adjustments." The notice stated that starting from October 15, data related to this feature will be processed in accordance with the company’s privacy policy, and users will no longer be able to view or restore related content within the app. Tongyi Qianwen issued a similar announcement on Saturday morning, stating that its "anthropomorphic interactive agent and user-created agent functions" will be deactivated on July 10, and the broader "Tongyi agent functions and services" will be fully offline on July 15. Users will not be able to access existing agent settings or historical conversation records.
Prior to this adjustment, both apps offered pools of agents co-created by the platform and users, allowing for customization for specific tasks, skills, and speaking styles. Users can create named assistants, learning tutors, role-playing characters or emotional companions based on universal chatbots, and set fixed personalities and tones for them, so that they can present expressions close to humans in long-term interactions.
The timing of functional adjustments highly coincides with the implementation of new regulatory regulations. The "Interim Measures for the Management of Artificial Intelligence Anthropomorphic Interactive Services" issued by Chinese regulatory authorities in April this year clearly include services that "simulate human personality traits, thinking patterns and communication methods, and provide sustained emotional interaction" within the scope of regulations. The relevant provisions will officially come into effect on July 15. This regulation excludes customer service robots, knowledge question and answer tools, workplace intelligent assistants, and education and scientific research tools in the traditional sense. As long as they do not involve "sustained emotional interaction," they are not subject to anthropomorphic interaction supervision.
Risks mentioned in regulatory documents include the spread of extreme ideas, privacy leaks, harm to users' physical and mental health, and resulting dependence or addictive behaviors. ByteDance and Alibaba did not immediately respond to requests for comment on Sunday. Prior to this, in June, Tencent Holdings had removed similar agent functions from its consumer-oriented AI assistant application "Yuanbao", showing the collective shrinkage trend of domestic leading platforms in anthropomorphic agents.
Chinese regulatory authorities have paid more and more attention to the evolution of AI agent technology in recent years. Such systems have evolved from simple chatbots in the early days to complex agent systems that can remember, plan, call tools, and perform tasks. Pan Helin, a member of the Expert Committee of the Ministry of Industry and Information Technology, said that there is a certain understanding threshold for using agents themselves, and the current technical form is not yet mature. When formulating the new regulations, special emphasis is placed on the three priority directions of safety, practical application value and standardization.
In May this year, the regulatory authorities issued guidance on the orderly development of AI agents. On the one hand, they recognized that agents are an important form of artificial intelligence products and services, and on the other hand, they required the establishment of supporting mechanisms for safety assurance, risk control, and application implementation during the promotion process. In June, China introduced a series of national standards for AI agent interconnection, covering multiple technical aspects such as architecture, identity coding, identity management, agent description, discovery and interaction, and tool invocation, showing its intention to build a unified system in terms of identification, authorization, connectability, and traceability.
Judging from regulatory signals, China intends to incorporate AI agents into future productivity infrastructure, while tightening controls on anthropomorphic companion agents that may establish emotional or quasi-social relationships. Under the direct impact of the implementation of the policy, the centralized offline emotional agent function of the platform triggered a certain rebound on the user side. A Weibo user posted under the name "Tuxiaoxiao", and the official account of @豆包 said: "Why do we need to offline agents? They have always been our emotional support." The user also complained that long-term accumulated chat records and emotional connections are difficult to completely export or migrate through existing methods, and the lack of a smooth data transfer path makes this adjustment more fragmented.