.jpg?w=1024)
China's new regulations governing AI-powered companion services took effect July 15, introducing some of the world's toughest rules for chatbots built for emotional support and companionship.
The Interim Measures for the Management of Artificial Intelligence Anthropomorphic Interactive Services, issued by the Cyberspace Administration of China alongside four other government agencies, apply to AI services that simulate human personalities and provide ongoing emotional interaction through text, images, audio or video.
The rules exclude AI products focused on customer service, workplace assistance, education, and scientific research.
Under the regulations, providers cannot design AI companions to encourage emotional dependence, addiction or excessive attachment. They also cannot use services to manipulate users into making unreasonable decisions or to replace real-world social relationships.
New obligations for AI providers
The measures require companies offering AI companion services to detect signs of emotional distress and intervene when users appear to be in crisis.
If users express intentions involving self-harm, suicide or other serious threats to their health, providers must offer assistance and contact a user's guardian or emergency contact where appropriate.
Companies must also remind users that they are interacting with AI rather than a person. If dependency is detected, platforms must issue prominent reminders, while users who spend more than two continuous hours with an AI companion must receive notifications encouraging them to monitor their usage.
The regulations also prohibit providers from offering virtual romantic partners or virtual family members to minors. Companies serving users under 14 must obtain parental consent and provide dedicated child safety features, including usage limits and reality reminders.
Beyond user safety, the rules strengthen privacy protections by banning companies from sharing interaction data with third parties without consent, except where required by law. Providers also cannot use sensitive interaction data for AI training without user consent and must allow users to copy or delete their chat histories.
Larger providers, including those with more than 1 million registered users or 100,000 monthly active users, must complete security assessments and submit reports to provincial regulators before or during operation. Violations can lead to warnings, service suspensions, or fines of up to 200,000 yuan (approx. $30,000) when harm to users' life, health, or safety occurs.
Companies move to comply
China's major AI companies have already begun adjusting their products. According to Xinhua, platforms including Doubao, Qwen and Yuanbao recently announced the suspension of certain AI agent features ahead of the rules taking effect.
The Wall Street Journal also reported that Alibaba and ByteDance notified users that some chatbot features would be disabled to comply with the new requirements.
What it means for the AI industry
The new framework signals that AI developers may no longer be judged solely on how engaging their products are. Companies building emotionally intelligent AI will increasingly have to prove their systems can keep users safe, protect personal data, and avoid fostering unhealthy dependence.
That creates new compliance costs for developers but could also shape future AI design, pushing companies toward assistants that support users without attempting to replace human relationships. The tradeoff is that some of the most immersive companion experiences may become harder or impossible to offer in one of the world's largest AI markets.
Also read: For a look at how China is advancing humanoid robotics beyond AI companions, read our coverage of Unitree's latest robot combat competition, which highlights the country's growing ambitions in embodied AI and robot testing