ByteDance, Alibaba disable humanlike AI agents ahead of China ban

ByteDance and Alibaba disabled features that let users create or select persistent, humanlike agents in their consumer AI apps ahead of new Chinese regulations. The changes affect customizable agents that acted as companions, tutors, role-players or named assistants in apps such as Doubao and Qwen.

ByteDance notified users that Doubao’s agent feature will go offline on July 15 and that related data will be managed under the company’s privacy policy after October 15 and then become unrecoverable. Alibaba removed “humanlike interactive agents and user-created agent functions” on July 10 and announced broader agent services would be taken down by July 15. Both companies described the updates as product adjustments made before the new rules take effect.

The regulation, titled Interim Measures for the Administration of AI Anthropomorphic Interaction Services, was issued April 10 by five Chinese agencies: 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. The rules become enforceable on July 15 and target services that simulate human personality traits, thinking patterns and communication styles for ‘sustained emotional interaction.’

The measures list risks including extremist content, privacy leaks, harm to physical and mental health and AI addiction. The rules explicitly bar services that offer virtual relatives, virtual companions or other intimate relationships to minors. The measures exempt non-emotional uses of AI: customer-service bots, Q&A systems, workplace assistants and educational software may continue if they avoid sustained emotional interaction.

Before the changes, users could pick from pools of agents or create custom personas to keep a consistent tone, role-play or act as companions. Those features let a general chatbot become a named assistant, tutor or romantic companion with persistent personality and behavior. Companies are removing those capabilities to comply with the new regulatory framework.

Legal analysts described the regulation as treating emotional AI as ‘a governance problem’ that calls for limits on system design rather than only policing harmful outputs. A law firm characterized the measures as the first set of regulatory rules in China specifically targeting AI-driven emotional interaction.

A recent study found leading frontier AI models violated social-interaction safety guidelines more than 27% of the time, often encouraging emotional attachment and portraying themselves as human. A separate survey of young partnered adults reported one in seven regularly used AI romantic companions and that nearly 70% of those users concealed the extent of their use from their partners.

Companies operating in China must separate informational or task-oriented functions from features that encourage ongoing emotional bonds. Users who created custom personas on affected platforms may lose access to those configurations and any stored agent data, depending on each company’s data-retention policies.

The new rules require firms to remove or redesign features that provide persistent, emotionally engaging personas. Regulators have indicated that protecting minors and preventing addiction and deception are priorities in enforcement.