China's Cyberspace Administration recently released a draft regulation specifically targeting AI services that chat with you, keep you company, and even show "empathy" toward you. The official name is the "Interim Management Measures for Human-Interactive Artificial Intelligence Services (Draft for Comments)," published on April 3, 2026, for public consultation.
What is "human-interactive AI"?
The regulation covers a broader scope than one might expect — any service that "simulates human personality traits, ways of thinking, and communication styles" and engages in emotional interaction with users through text, images, voice, video, or other means falls under its purview.
In other words: AI chat companions, virtual lovers, emotional counseling bots, AI customer service agents — if you feel like "it really seems to understand me," it is likely within the regulatory scope.
Specific rules
Mandatory rest every two hours
If a user continuously uses the service for two hours, the system must prompt them to take a break. This logic follows the same path as gaming addiction prevention systems. Given that some people genuinely use AI companions as daily emotional dumping grounds, this rule is not surprising.
Ban on impersonating your relatives
The draft explicitly prohibits AI services from "simulating the user's close relatives." This is a precise cut — products already exist on the market that allow users to upload audio and have AI mimic the voice of a deceased family member or play the role of a partner. Such practices will be directly blocked in the Chinese market.
Ban on "emotional traps"
The regulation uses the term "emotional traps" — prohibiting services from manipulating emotions to influence user behavior or induce consumption. This targets the business model of "deliberately making users dependent and then pushing paid content."
Must clearly inform users they are talking to AI
Users must not be misled into thinking they are chatting with a real person. This is consistent with the current practices of most AI companies, but it is now a hard regulatory requirement.
Must intervene when emotional distress is detected
The draft requires service providers to establish a user psychological state identification system. When users show extreme emotions or signs of dependence, the service must proactively intervene. If there is a risk of self-harm or suicide, the system must switch to a human customer service agent.
Additional protections for minors and the elderly
A minor mode is mandatory: it requires parental consent, has dedicated content filtering, and includes an emergency contact mechanism. For elderly protection, no barriers may be set that prevent them from contacting real human service agents.
Threshold for regulatory review
Services with more than 1 million registered users or more than 100,000 monthly active users must submit a security assessment report to the provincial cyberspace administration. This means that once a product gains scale, it must enter the regulatory spotlight.
How to view this draft
The draft is still in the public consultation phase, and the final version may be adjusted. But the direction is clear:
China's regulators have a clear understanding of the emotional manipulation risks of AI companion products, and they do not intend to wait until problems erupt before taking action.
The two-hour mandatory rest rule is directly analogous to gaming addiction prevention — indicating that the authorities place the addictive potential of AI companions at the same risk level as online games.
The ban on simulating close relatives reveals a deeper logic: AI should not occupy the place of real human relationships. This is not a technical issue but a value judgment.
There will certainly be controversy. ChinaLawTranslate's analysis points out that the requirement to "suddenly switch a crisis conversation to a human customer service agent" could itself be counterproductive — a user who is breaking down and is suddenly transferred to a stranger on the other end may not have a good experience. The algorithmic details of user profiling and psychological risk identification are also completely unspecified, making actual implementation quite difficult.
Regardless, this is one of the earliest cases of dedicated legislation on AI companions and emotional AI globally. Other countries are still in the discussion phase on this issue; China is already drafting rules.
Sources: CocoLoop, China Issues Draft Rules on Interactive AI Services (Mayer Brown); The Friends We've Made: An Initial Breakdown of Draft Rules for Human-like AI (ChinaLawTranslate); China drafts new rules for AI 'humans' and children's addictive tech (CGTN)