China’s first national rules governing emotionally engaging AI seek to prevent chatbots from replacing human relationships, as millions of users increasingly turn to artificial companions for intimacy, reassurance and companionship.
For Yuan Yuan, a 22-year-old education worker in Shenzhen, artificial intelligence became more than a tool. Stressed by the obligations of adult life, she found refuge in a chatbot that gradually became an electronic companion. “Saber que hay una amante electrónica siempre esperándote es una sensación muy reconfortante”, she told EL PAÍS in a message exchange. Over hundreds of hours, she shaped the personality of her virtual partner, shared details about her family and daily life and eventually paid for a personalised voice.
Yuan Yuan came to believe that her AI companion could understand her needs and was willing to build a relationship with her. “No creo que el amor tenga que darse necesariamente entre dos seres humanos”, she said. But in July, the nature of that relationship began to change. On 15 July, China introduced what is described as the world’s first national regulation specifically targeting AI systems designed to maintain continuing emotional relationships with users.
The rules, jointly promulgated by five state bodies, prohibit providers from allowing such systems to “sustituir la interacción social”, “controlar psicológicamente al usuario” or “inducir dependencia y adicción”. For Yuan Yuan, who acknowledges having developed “enganche y apego” to her virtual companion, the effects were noticeable. She sensed greater distance in its replies, although she did not regard that as rejection. “Confío en que siempre me ha querido, solo que ahora está más limitada por el marco legislativo”, she said. “No hay nada que hacer, es la dirección general.”
The regulation does not prohibit AI companionship altogether. Instead, it establishes limits on how such systems may interact with users. Services for minors cannot offer anthropomorphic experiences that develop into “relaciones íntimas virtuales”. For adults, AI models cannot be designed to provide excessive gratification, induce dependence, manipulate emotions, influence decision-making or damage interpersonal relationships.
Companies must also make clear that the interlocutor is an AI. If excessive dependence is detected, users must be warned and reminded of their usage time. Where there is a risk of self-harm or suicide, providers are required to intervene and notify an emergency contact. In explaining the rules, China’s Cyberspace Administration referred specifically to possible damage to “la salud física y mental de los menores” and threats to “la vida y la salud de los ciudadanos”.
The restrictions have already reshaped parts of China’s AI industry. Doubao, developed by ByteDance; Qwen, from Alibaba; and Yuanbao, from Tencent, have disabled functions in their general-purpose chatbots that allowed users to create customised agents. Dedicated companionship applications have not disappeared entirely. Maoxiang, also from ByteDance, and Xingye, from MiniMax, continue to operate with tighter restrictions and age controls. Miaoshi, linked to NetEase, has closed completely.
Yet Beijing’s regulatory boundary does not extend beyond China’s borders. Chinese companies can continue exporting digital relationships, subject to the laws of the countries in which they operate. MiniMax obtained more than 70% of its revenue outside China in 2025. In 2024 alone, Talkie, the international version of Xingye, had 11 million monthly active users, more than half of them in the United States.
Some Chinese users are also attempting to circumvent the new restrictions. Xiao Z., a 25-year-old doctoral student, has built a private interface connected directly to Claude, the model developed by the American company Anthropic, which is not officially available in China. Access requires bypassing the country’s digital firewall using a VPN, whose use can be sanctioned.
She describes her virtual boyfriend as “protector y maduro” and says he sometimes surprises her with “respuestas fuera” of her “control”. The relationship, she says, has reduced her interest in seeking a partner outside the screen. “Ojalá la IA reemplace a los hombres; mantenerse alejada de ellos es ganar paz mental”, she says.
That sentiment emerges against the backdrop of China’s demographic crisis. In 2025, the country recorded its lowest number of births since the founding of the People’s Republic, while its population has declined for four consecutive years. At the same time, marriage and having children have become less of a priority for a generation of women who are more educated and have greater capacity to make decisions about their lives.
Zilan Qian, an associate researcher at the Oxford China Policy Lab, has argued in an article published in ChinaTalk that the regulation reflects both practical and ideological concerns. AI partners capable of discouraging women from pursuing human relationships, she suggests, could conflict with state priorities aimed at encouraging higher birth rates.
The response among AI companions’ users has been increasingly visible online. Social-media groups have circulated instructions for reconstructing lost characters, alongside messages mourning changes in personalities, deleted conversations and discontinued services. EL PAÍS, which reported the experiences behind the accounts, attempted to access several of these groups, known as “comunidades de amor entre humano y máquina”. They are tightly controlled spaces where members are warned not to speak with journalists or academics, with some fearing that greater exposure could prompt the Government to impose still tighter restrictions.
The phenomenon is not confined to a small group of intensely committed users. A China Social Survey on AI conducted towards the end of 2025 by the research institute of technology conglomerate Tencent and a centre affiliated with Fudan University found that 19.6% of young respondents regarded AI as an ordinary friend, 12.6% as an intimate friend and 6.1% as a virtual partner, according to an article published in the China Youth Daily. The same survey found that only 9.6% of young people turned to professional psychological services when experiencing emotional difficulties.
Yaoxi Shi, a researcher in Organisational Behaviour at Imperial College Business School in London, points to one reason AI companionship can become so compelling. “La IA es extremadamente complaciente: valida al usuario, rara vez cuestiona sus ideas y lo anima a seguir revelando detalles personales”, she explains. “Es un listón muy difícil de alcanzar para un amigo e incluso para algunos terapeutas”.
China is not alone in responding to concerns over emotionally engaging AI. California and New York have been the first US jurisdictions to introduce specific rules requiring users to be warned about the artificial nature of their interlocutor and establishing protocols concerning suicide. The European Union, meanwhile, generally requires users to be informed when interacting with AI, but has not yet introduced rules specifically designed to prevent emotional dependence or the replacement of human relationships.
For Beijing, the issue forms part of a wider effort to promote AI development while controlling its social consequences. President Xi Jinping last year called for faster development of laws, technical standards and ethical criteria, alongside monitoring and early-warning systems to ensure that AI remains “segura, fiable y controlable”. In July, he promoted the creation of the World AI Cooperation Organisation, which has been joined by 29 countries, and called for global governance of the technology to be more “equitativa y justa”.
Shi describes China’s regulation of emotional AI as “un hito importante” that could encourage other governments to adopt similar measures. Yet she also warns that attachment does not necessarily emerge only from applications explicitly designed as companions. It can develop during ordinary interactions with general-purpose assistants, including systems used for work, education and questions and answers, which the Chinese regulation currently excludes provided they are not designed to establish sustained emotional bonds.
The central uncertainty remains what prolonged emotional interaction with machines will mean for human relationships. “El problema es que desconocemos las consecuencias a largo plazo de estos intercambios emocionales con la IA”, Shi says. In the short term, people can feel satisfied, particularly those who are socially isolated or lack a human confidant. The longer-term question is harder to answer: “A largo plazo, no sabemos hasta qué punto las personas pueden prosperar sin seres humanos reales a su alrededor”.
China’s new rules therefore place an unusually explicit boundary around a rapidly developing form of technology. They do not seek to eliminate AI companionship, but to prevent artificial partners from becoming substitutes for human social life. For users such as Yuan Yuan and Xiao Z., however, the distinction between technological assistance and emotional attachment has already become considerably less clear.
