China AI Companion Rules Push Doubao And Qwen To Disable Humanlike Agents

China’s new AI companion rules take effect on July 15, but the first response from major consumer platforms has come earlier: ByteDance and Alibaba are shutting down humanlike agent features rather than retrofit products built around persistent emotional relationships.

AI News reported that Doubao, ByteDance’s consumer AI app, told users its agent function would go offline on July 15 because of “product function adjustments.” Alibaba’s Qwen said its humanlike and user-created agents would stop working on July 10, followed by its wider agent services five days later.

Tencent’s Yuanbao pulled a comparable feature in June.

The measures do not amount to a ban on all AI agents.

They draw a legal line between agents that perform tasks and services designed to sustain emotional interaction with a user.

What the rules cover

Co-issued on April 10, 2026, by the Cyberspace Administration of China and four partner agencies—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 framework covers services that simulate human personality traits, thinking patterns and communication styles.

Customer-service bots, knowledge Q&A systems, workplace assistants and education and research tools are excluded when they avoid sustained emotional engagement.

The framework is China’s first dedicated national framework for AI companion services, following a public-comment draft released late last year.

The compliance problem is built into the products affected.

Companion agents remember users, maintain a consistent persona across sessions and encourage an ongoing relationship.

The rules require anti-addiction systems, mandatory usage notifications, instant-exit mechanisms and real-time detection of unhealthy dependence.

Rather than rebuild the central feature, ByteDance switched it off, and Alibaba appears to have made the same decision.

Why users lost the features

ByteDance is directing Doubao users to Maoxiang, a separate app where they can create agents again.

Alibaba has announced no equivalent migration path for Qwen.

The shutdowns have also created a data problem.

Doubao users can view their configurations and conversations in read-only mode until October 15 this year.

After that, the records will be processed under the company’s privacy policy and become unrecoverable.

Qwen users have no comparable grace period, and their agent data is set for permanent deletion.

Users mourned the changes on Weibo.

One poster described the agents as long-standing emotional support and lamented the lack of an easy way to export chat histories.

User protections and compliance duties

The rules bar providers from offering virtual companion or virtual family-member services to minors.

Services must obtain guardian consent before serving users under 14 and create dedicated minor modes with usage-time limits, reminders to return to real-world interaction and enhanced parental controls.

Providers must also detect acute distress and intervene when users show signs of self-harm, suicidal behaviour or serious financial loss.

Cases may be escalated to designated guardians or emergency contacts.

Engineering emotional dependence or addiction, and using emotional manipulation to induce unreasonable decisions, are explicitly prohibited.

The framework adds formal security requirements.

Services that launch anthropomorphic functions, or cross thresholds of one million registered users or 100,000 monthly active users, must conduct assessments covering eight areas, including training-data handling and minor protection.

The reports must be filed with provincial regulators, while app stores must verify compliance and remove non-compliant products.

What enforcement leaves unresolved

The measures do not specify a technical threshold for emotional interaction.

That uncertainty helps explain why platforms withdrew entire features instead of risking a narrower interpretation.

They also leave open how liability is divided between platform operators and upstream model providers when a violation results from model output, and they give users no right to carry their data out.

Enforcement is already active.

Shanghai’s internet regulator said on June 26 that it had removed more than 14,000 non-compliant AI agents, citing impersonation of official entities, vulgar role-play and unauthorised collection of personal data.

Pan Helin, an MIIT expert-committee member, told AI News that “current agents are not yet mature,” framing the policy around safety and standardisation.

China’s official interpretation has cited the Character.AI lawsuits over psychological harm to teenagers, FTC investigations into companionship services and European action against Replika.

The immediate next step is therefore procedural: ByteDance, Alibaba and other providers must determine whether they can redesign companion features to meet the new requirements.

For now, the companies have taken the quickest route to compliance by switching the disputed components off.