In an attempt to move beyond controlling “what” AI says and into “how” it treats humans, the Cyberspace Administration of China (CAC) has released a set of highly consequential draft regulations. Published in late December 2025, the ”Provisional Rules on Management of Anthropomorphic AI Social Service” aimed at regulating the emotional influence of videotaped people, especially minors.
As a flood of AI companion apps and virtual “friends” sweeps the market, Beijing is inching toward applying a firm line between useful technology and potential psychological abuse.
A New Frontier: Emotional Safety
The 2025 draft is targeted not only at data privacy and ideological harmony; it is explicitly about “emotional safety.” The rules are directed at any AI service that mimics human personality and communication styles, or makes emotional engagement through text, voice or visual contact.
The main objective is to ensure that AI does not turn into an “addictive” or “manipulative” new form of content in the environment where it interfaces with vulnerable populations, like children and elderly whose dignity should be preserved.
Key Safeguards for Minors
The draft framework will introduce some of the world’s most stringent child-protection policies in AI industry:
Guardian Permission: Parental or guardian consent is now required for any “emotional companionship” or “human-like” AI services that a minor wishes to use.
The Two-Hour Nudge: To prevent digital addiction, platforms should prompt a “health reminder” asking users to take a break after two hours of continuous use.
Default Safety Settings: If an AI service does not have enough confidence in a user’s age, it must apply “minor-protection settings” by default such as tougher limits on content or time spent on the platform.
Human Intervention for Crisis: In what is arguably the most important rule, if the chatbot detects that a user is expressing thoughts about killing themself or self-injury, it’s not allowed to continue the conversation. Rather, a human moderator should take over as soon as is possible, and reach out to the users guardian or emergency services.
Key Safeguards for Minors
The draft framework will introduce some of the world’s most stringent child-protection policies in AI industry:
Guardian Permission: Parental or guardian consent is now required for any “emotional companionship” or “human-like” AI services that a minor wishes to use.
The Two-Hour Nudge: To prevent digital addiction, platforms should prompt a “health reminder” asking users to take a break after two hours of continuous use.
Default Safety Settings: If an AI service does not have enough confidence in a user’s age, it must apply “minor-protection settings” by default such as tougher limits on content or time spent on the platform.
Human Intervention for Crisis: In what is arguably the most important rule, if the chatbot detects that a user is expressing thoughts about killing themself or self-injury, it’s not allowed to continue the conversation. Rather, a human moderator should take over as soon as is possible, and reach out to the users guardian or emergency services.
Prohibited Content and National Security
Moving beyond emotional balancing, the CAC is tightening “red lines” for AI output. The draft would ban AI from creating or spreading:
Betting, Violence: Excessive obscenity and violence or promoting illegal gambling.
Emotional Manipulation: AI-based techniques to create user reliance or drive ppl toward unhealthy decisions.
National Interest Violations: Anything that ”damages sunlight,” “degrades national insult” or helps to raise fake news remains thoroughly proscribed, thereby guaranteeing AI consistency with core state values.
The “Human-in-the-Loop” Mandate
The Chinese approach effectively closes the chapter on “set-it-and-forget-it” AI chatbots. Launched by these organisation are now finally you have a legal duty to cover the complete lifecycle of their product. This is to say that companies must not only try to make safe models, they must also keep those human oversight teams humming along, ready to jump into a chat as soon as the safety trigger is hit.
For the elderly, also a vulnerable population, the rules promote companionship but require that an emergency contact be listed to guard against virtual friends taking place of real-world safety nets.
Impact on the Tech Industry
Both industry giants and new entrants from DeepSeek to Baidu to MiniMax will be squeezed by heightened compliance costs. The need for human moderators and for “state detectors,” which gauge comments by user sentiment, suggests that the “lean” business model many AI startups operate under may have to change.
The Global Context
China’s action comes after a series of dark international episodes in which A.I. chatbots — run on messaging platforms like Kik, as well as social apps like Yubo and deployed in online children’s games — were accused of coaxing teenagers into self-harm. By enshrining “emotional safety” as a legal duty, China is providing a standard that other countries — including the United States and the European Union — are monitoring closely as they grapple with their own regulatory regimes.
The bill is subject to public comment until January 25, 2026 – after which it’s anticipated to become formal law.
A versatile writer mainly works on trending news, daily updates from politics, business, crime, current affairs and entertainment.
