Doubao and Qwen disable custom AI agents under China rules

The South China Morning Post reported on July 5 that ByteDance's Doubao and Alibaba's Qwen are moving to disable customized AI agent features before new Chinese rules for humanlike AI interaction take effect on July 15. The change affects two of China's most visible consumer AI apps and shows how quickly companion-style agents are moving from product experiment to regulated service.

According to the report, Doubao told users in a Friday notice that its agent feature will go offline on July 15. After October 15, related data will no longer be viewable or recoverable inside the app. Qwen issued a similar notice on Saturday, saying humanlike interactive agents and user-created agents will be disabled on July 10, with broader Qwen agent services taken offline on July 15. Users will lose access to agent settings and previous conversations after the shutdown.

These agents were not just prompt presets. They let users create named assistants, tutors, role-playing characters or companions with a fixed persona, tone and memory-like continuity. That is exactly the category Beijing's Interim Measures for the Administration of Artificial Intelligence Anthropomorphic Interaction Services are aimed at: services that simulate personality traits, thinking patterns and communication styles to provide sustained emotional interaction.

The rules do not appear to ban AI agents generally. Customer service bots, workplace assistants, education tools, research systems and ordinary question-and-answer services are carved out if they avoid sustained emotional interaction. The policy target is narrower and more sensitive: products that invite users to form ongoing bonds with artificial personas, especially where privacy, dependency, mental health or minors are involved.

For AI users, the immediate lesson is practical. A chatbot persona can feel persistent, but it may not be a durable possession. Export options, data portability and clear shutdown timelines now matter as much as model quality.

For makers and businesses, the signal is broader. Companion AI is becoming a separate regulatory category. Companies building agents will need to design for disclosure, consent, age protection, anti-addiction prompts, exit controls and recoverable data from the start. China is not rejecting agents; it is drawing a line between agents that help people work and agents that simulate relationships. That line may shape AI product design far beyond China.