AI agents: Beijing imposes its rules, the Chinese giants obey

China is forcing ByteDance and Alibaba to deactivate certain functions of personalized AI agents before new rules come into force on July 15, 2026. Beijing wants to regulate services capable of simulating a human personality, an intimate relationship or a prolonged emotional interaction.

AI: Beijing closes the door to agents that are too human

New Chinese regulations target AI services that imitate human traits. It concerns AI agents capable of adopting a stable personality, an emotional tone or a role close to the virtual companion. Beijing no longer wants to let these products evolve without safeguards.

ByteDance announced the deactivation of custom agents on Doubao starting July 15. The associated data may become unrecoverable after October 15, depending on the information sent to users.

Alibaba is going even faster. Qwen must remove certain human-looking interactive agents starting July 10, before a broader shutdown on July 15. The two giants prefer to obey quickly rather than risk a standoff with the authorities. The signal is clear. In China, emotional AI is no longer just a product innovation. It becomes a subject of governance, mental health and social protection.

ByteDance and Alibaba adjust their products

Doubao and Qwen offered agents that users could customize. Some served as assistants, tutors, or role-playing characters. Others took the form of virtual companions with a consistent conversational style.

It is precisely this emotional continuity which worries Beijing. A chatbot that answers a question remains authorized. An agent designed to create a lasting relationship, mimic attachment, or replace a human connection enters a much more sensitive area.

Work tools, customer services, educational assistants and question-and-answer software are not directly targeted. The boundary is in the intent of the service. Productive AI remains acceptable. Relational AI becomes strictly monitored.

This distinction also protects large Chinese companies. Beijing is not blocking all innovation. It draws a limit around uses deemed psychologically risky, especially when minors are concerned.

Minors at the center of the system

The new rules impose strong restrictions on services that offer virtual parents, virtual companions or intimate relationships to minors. The text also mentions the risks of dependence, invasion of privacy and negative effects on mental health.

This approach responds to growing concern. AI agents can learn a user's habits, adapt their tone, and encourage continued interaction. The more human the system appears, the more the risk of attachment increases.

China wants to prevent these products from becoming uncontrolled emotional substitutes. It therefore requires platforms to review the very design of the services, not just the content generated. This is where the decision becomes important. Beijing is not just asking for more transparency. The country directly limits certain interaction patterns. For Chinese AI giants, compliance means removing entire functions.

This decision could slow down certain general public uses, but it does not kill the market. Professional, educational, and productive AI agents will continue to exist. What is disappearing is the idea that an AI can be sold as a human connection without serious oversight.

The Chinese giants therefore obey because the issue goes beyond simple compliance. In China, AI must remain useful, supervised and aligned with the priorities of the State. The removal of custom agents shows that the next battle won't just be about template power. It will focus on the place that these models have the right to take in daily life, at a time when several countries are already starting to think about the governance of agents.