News
China Forces Doubao and Qwen to Shut Down AI Companions Ahead of New Law

ByteDance and Alibaba are simultaneously phasing out personalized, humanlike AI agent features in Doubao and Qwen ahead of China's new anthropomorphic interaction regulations, which take effect on July 15, 2026.
ByteDance and Alibaba have simultaneously announced they are shutting down the ability to create personalized AI agents in their most popular chatbots, Doubao and Qwen respectively. The move lands on the very day new Chinese rules governing anthropomorphic AI services take effect, underscoring how quickly Beijing can force changes on the country's largest tech firms.
Overnight between Friday and Saturday, Doubao sent users a notification that its agent feature would stop working due to a product feature adjustment. Qwen's Saturday announcement read almost identically: the platform said its humanlike interactive agents and user-created agent features would be disabled on July 10, with Qwen's broader agent features and services shutting down on July 15.
Beijing's new rules
Both decisions stem from the Interim Measures for the Administration of Artificial Intelligence Anthropomorphic Interaction Services, which take effect on July 15. The rules cover AI services that simulate human personality traits, thinking patterns, and communication styles in order to sustain ongoing emotional interaction with a user.
Chinese regulators justify the new rules by citing concerns over the spread of extreme content, privacy violations, harm to users' mental and physical health, and the risk of addiction to conversations with bots posing as humans. The law does not cover customer service bots, question-and-answer systems, work assistants, or educational and scientific tools, provided they do not engage in sustained emotional interaction.
How the companies responded
Both companies limited their public statements to technical language about product feature adjustments, avoiding any direct link between the decision and the incoming law. Doubao and Qwen are advising users to back up important content in advance, via screenshots or text exports, before agent configurations and chat histories disappear for good.
On Doubao, users will still be able to view the settings of agents they created and their chat history during a transition period, but after October 15 the data will no longer be accessible and cannot be recovered. Qwen works on a similar basis, except the process is split into two stages five days apart.
The bigger picture
This is the latest step in an oversight architecture for generative AI that Beijing has been building for months, alongside earlier requirements to label synthetic content and register models before public release. China is regulating the AI sector faster and more directly than the European Union or the United States, where debate over AI companions and their impact on mental health is only now picking up pace following a string of high-profile lawsuits.
China's market for persona chatbots has exploded over the past two years, driven by apps that let users create custom characters for hours-long, emotionally engaged conversations. It's this category, not general-purpose office chatbots or coding assistants, that the new rules target.
For Doubao, one of China's most popular chatbots, and Qwen, Alibaba's flagship consumer product developed alongside the open-source model family of the same name, losing agent features means losing a slice of user engagement right before the summer holiday period, when traffic to entertainment apps typically rises.
Neither company has said whether or when agent features might return in a revised, compliant form. Market observers expect other Chinese platforms with similar AI companion features to follow Doubao and Qwen's lead before July 15, in order to avoid penalties for non-compliance with the new rules.
For European observers of the AI market, this is a signal that regulation of persona chatbots and AI companions could quickly become its own legal category, separate from general rules on high-risk AI systems. The EU's AI Act does not currently carve out an explicit category for anthropomorphic interaction, but a growing number of reports about user addiction to AI companions could accelerate similar discussions in Brussels.
Sources: South China Morning Post (scmp.com), TechNode (technode.com).


