ByteDance and Alibaba are shutting down user-created AI agent features in Doubao and Qwen ahead of new Chinese rules governing humanlike AI interaction services.
Doubao and Qwen both notified users that agent-creation features will be discontinued before or on July 15, when China’s Interim Measures for the Administration of Anthropomorphic AI Interaction Services take effect. After the shutdown, users will no longer be able to create new agents, and existing user-created agents will stop functioning. The platforms are giving users a transition period to view or export agent configurations and chat histories before related data becomes inaccessible.
The affected features are the consumer-facing tools that let users create customized AI personas or agents for ongoing interaction. The core chatbot, productivity, search, and question-answering functions are not being withdrawn. The new rules are aimed specifically at services that simulate human personality traits, communication styles, or sustained emotional interaction, rather than ordinary workplace assistants or knowledge tools.
The Cyberspace Administration of China issued the rules with four other agencies in April. The measures require providers of anthropomorphic AI services to manage risks around emotional dependency, psychological harm, minors, personal data, and AI-generated content. They prohibit services from encouraging users to replace real social interaction, inducing addiction or emotional dependence, or manipulating users into unreasonable decisions.
The rules also impose operational requirements that are difficult to apply to large pools of user-generated agents. Providers must identify risks during interactions, intervene in extreme situations, offer reminders after more than two hours of continuous use, provide ways for users to exit interactions, and run safety assessments for large-scale services. Services aimed at minors face stricter limits, including a ban on virtual relatives and virtual companions for underage users.
The shutdown fits into a broader Chinese regulatory move from model capability control to product-level control. Beijing is already considering restrictions on overseas access to frontier Chinese AI models. The Doubao and Qwen changes show the domestic side of the same trend: Chinese platforms can continue scaling AI products, but higher-risk uses are being moved into narrower, more reviewable product structures.
User-created agents are difficult to govern because the platform does not define the persona, purpose, tone, or interaction pattern in advance. A company can approve its own assistant or customer-service bot against a compliance checklist. It is much harder to guarantee that millions of user-built agents will not become companions, therapists, role-play partners, or manipulative emotional products.
That governance problem is showing up in different forms across the AI market. Enterprise vendors are building controls for agent identity and authorization, while standards groups are trying to define how agents prove who they are and what they are allowed to do. China’s consumer AI rules approach the problem from another direction: when agents become humanlike, emotionally persistent, and user-defined, the platform is responsible for keeping the interaction inside regulatory bounds.
Sources: TechNode, Cyberspace Administration of China, 21st Century Business Herald
–
By the Control Plane Editorial Team