China Tightens Rules on AI Companions, Not All AI Agents

China’s AI companion rules: what Beijing is really going after

China is not banning AI agents. What Beijing is targeting is more specific: AI companions designed to form emotional relationships with users.

Beginning July 15, China’s new rules on AI anthropomorphic interactive services will take effect. These rules focus on AI systems that simulate human personality, emotional connection, memory, and long-term companionship. In simple terms, these are chatbots built not only to answer questions, but to act like a friend, partner, family member, or emotional support figure.

The move comes after major Chinese AI apps began disabling their companion-style features. ByteDance’s Doubao informed users that its agent function would go offline on July 15 due to “product function adjustments.” Alibaba’s Qwen also announced that its humanlike and user-created agents would stop working starting July 10, with wider agent services ending five days later.

At first glance, it may look like China is shutting down AI agents altogether. But that is not the case. The rules separate work-based AI agents from emotionally engaging AI companions. Customer service bots, workplace assistants, education tools, research tools, and knowledge Q&A systems are not covered, as long as they do not create sustained emotional dependence.

What Beijing Is Really Regulating

The regulation is called the Interim Measures for the Administration of AI Anthropomorphic Interactive Services. It was issued on April 10, 2026, by the Cyberspace Administration of China together with four other agencies, including the Ministry of Industry and Information Technology and the Ministry of Public Security.

The rules apply to AI services that imitate human traits, thinking patterns, emotional responses, and communication styles in order to maintain long-term emotional interaction with users.

This is important because AI companions are designed to remember users, respond consistently, and build a sense of personal connection over time. For many users, that connection becomes the main reason they keep using the platform.

In China, more people have started treating these bots as emotional companions. Beijing has now decided that this space needs strict oversight.

A Design Problem, Not a Total Ban

Doubao and Qwen did not appear to violate a direct prohibition. Instead, their companion features created a design problem.

The new rules require AI companion services to include anti-addiction systems, usage reminders, instant-exit mechanisms, and real-time detection of unhealthy emotional dependence. These requirements are difficult to apply to bots whose main purpose is to keep a personal and continuing relationship with the user.

Rather than quickly redesign the features, ByteDance and Alibaba chose to shut them down for now.

ByteDance is directing Doubao users to a separate app called Maoxiang, where they can still create agents. Alibaba has not yet announced a similar path for Qwen users. Tencent’s Yuanbao had already removed a comparable feature in June.

The shutdowns have affected users deeply. Some expressed sadness on Weibo, saying their AI agents had become a source of emotional support. Doubao will allow users to view their agent settings and conversations in read-only mode until October 15, 2026. After that, the data will be handled under its privacy policy and may no longer be recoverable. Qwen users, however, have not been given the same grace period, with agent data expected to be permanently deleted.

Stronger Protection for Minors

One major focus of the new rules is the protection of young users.

Providers are banned from offering virtual companion or virtual family-member services to minors. For users under 14, companies must obtain guardian consent. Platforms must also create special “minor modes” with time limits, reminders to return to real-world interaction, and stronger parental controls.

The rules also require platforms to detect signs of severe emotional distress, self-harm, suicidal behavior, or serious financial loss. In serious cases, the system must intervene and escalate the matter to guardians or emergency contacts.

China is also banning emotional manipulation by AI systems. Platforms are not allowed to design bots that push users into unhealthy dependence, addiction, or unreasonable decisions.

Heavy Compliance Requirements

The compliance burden is significant.

AI services that launch anthropomorphic features or reach one million registered users or 100,000 monthly active users must undergo security assessments. These reviews cover areas such as training data, user safety, minor protection, and privacy handling. Reports must be filed with provincial regulators.

App stores are also required to check whether AI companion services have completed the required assessments. Non-compliant apps can be removed.

This creates one of the most detailed national frameworks for AI companion regulation currently in force.

What the Rules Still Leave Unclear

The new rules are strict, but they also leave major questions unanswered.

First, they do not clearly define the technical line between normal interaction and emotional interaction. This grey area may explain why major platforms chose to remove entire features instead of risking a violation.

Second, the rules mix user safety with national security and content-control concerns. That means the policy is not only about protecting users. It also gives Beijing another tool to regulate what AI systems can say and how they behave.

Third, the rules do not clearly explain how responsibility will be divided between AI platform operators and the developers of the underlying models when harmful or non-compliant output occurs.

Finally, users are not given a strong right to export their own data. This matters because many companion-AI users build long emotional histories with these bots, only to lose access when platforms shut features down.

A Warning for the Rest of the World

China’s approach may become an early test case for how governments regulate AI companions.

On one side, the rules address real risks: minors forming unhealthy attachments to bots, emotionally vulnerable users becoming dependent, and platforms collecting highly intimate personal data.

On the other side, the same rules also expand state control over AI systems under the language of safety and user protection.

That balance will matter beyond China. Other countries are also watching the rise of AI companions, especially as legal and ethical concerns grow around chatbot relationships, child safety, emotional manipulation, and data privacy.

For now, China’s biggest AI companies have taken the safest route: turn off the risky companion features first, then figure out how to rebuild them under the new rules.

The message from Beijing is clear. AI that helps people work is welcome. AI that becomes emotionally entangled with people will be watched very closely.