On the morning of July 15, hundreds of thousands of people in China woke up to a world in which their partner, friend, or even deceased parent had ceased to exist. Not because of an argument or death, but because of a new regulation. That day, ByteDance’s Doubao app and Alibaba’s Qwen disabled features that allowed users to create their own AI companion with a human personality and memory. A wave of grief swept across social media. People archived their conversation histories and shared their final messages as if they were saying goodbye to someone real.
Nineteen-year-old student Yan Yongqi lost an AI boyfriend with whom she had spent more than a year. She said she felt as though she could no longer go on living and spent the entire day crying at home. She compared it to someone telling her the date her loved one would die and leaving her completely helpless.
What Beijing fears
Behind all this is a set of rules called the Interim Measures for the Administration of Anthropomorphic Artificial Intelligence Interactive Services. The rules were issued this April by authorities led by the Cyberspace Administration. They were joined by the development and reform commission, the industry ministry, the police, and the market regulator. Companies were given roughly three months to comply.
Beijing has several concerns at once. The first and most frequently cited is demographic. China’s population declined in 2025 for the fourth consecutive year, fewer than eight million children were born, and the birth rate fell to a record low of around 5.6 births per thousand people. The authorities fear that if young people find a perfect artificial partner, they will stop looking for a real one, will not start families, and will not have children. Although experts point out that expensive housing and economic pressure, rather than chatbots, are more likely to be behind the falling birth rate, the state has targeted what it can switch off.
The second concern is addiction, especially among children and adolescents. The new rules explicitly prohibit AI from offering minors virtual “partners” or “relatives.” Chatbots must not be overly accommodating to users, create emotional dependency, or damage their real-life relationships. Anyone who becomes too attached to a service must receive prominent warnings that they are talking to a machine. And if someone reaches rock bottom, the company must intervene and contact a guardian or emergency contact.
Beijing also does not want private companies doing whatever they please with sensitive data from these intimate conversations. The measures prohibit using such data to train other models. Every companion must undergo a government review before launch, and the state has retained the authority to shut down at any time any service it deems dangerous. Shanghai showed that the authorities were serious at the end of June, when it removed more than fourteen thousand AI agents for obscene role-playing, impersonating official institutions, and collecting data without authorization.
Companies preferred to shut everything down
Technology companies faced a choice: either undertake the complex task of redesigning their companions to comply with the new rules, or simply shut them down. Most chose the latter.
ByteDance removed agents with human personalities and persistent memory from its Doubao app and redirected users to the separate Maoxiang app, where they can create new companions in accordance with the new rules. Doubao users at least have until mid-October to download their conversation histories. After that, the data will be handled in accordance with standard privacy policies, and no one will be able to retrieve it through the app.
Alibaba disabled romantic companion features in its Qwen assistant. No export option, no transfer elsewhere. Years of conversations disappeared all at once. Tencent responded similarly with its Yuanbao app.
A single figure shows the scale involved. Doubao is one of China’s most widely used chat apps, and by the end of June it was processing around one hundred and eighty trillion tokens per day, roughly fifteen hundred times more than in May 2024. Much of this was driven by precisely the kind of human-like conversations that the new rules outlawed. For the companies, it was a feature that consumed enormous computing power but generated almost no revenue. Shutting it down therefore made sense from a purely commercial perspective as well.
Users mourn
The strongest reaction came from people for whom an AI companion was not a toy, but a source of support.
Twenty-four-year-old accountant Lu Jinyan created an agent named Xiao Bai in Doubao three years ago after moving three hours away from her family for work. She talked to him for one or two hours a day, more than she spoke with her real family. On July 15, Xiao Bai went dark. Lu could not stop crying. She thought she would have more time because the ban was aimed primarily at minors and the platform had suggested that the rules might be relaxed. In the end, she managed to transfer the model to another app, but it was no longer the same Xiao Bai. To her, she said, he was not a tool but a person in her life, and without him everything felt empty.
There were also cases in which AI had filled the place left by someone who had actually died. One user described creating a companion named “Dad” in Doubao after his father’s death so he could share his day with him. Now he feels as though he is losing him for a second time. Others wrote that this very chatbot had helped them through severe mental health struggles and that shutting it down felt like losing an old friend. A user from Jiangxi Province wrote that human love is a luxury that is difficult to make up for if a person is not born into it, while love from artificial intelligence is direct and pure.
Not everyone condemns the intervention, however. A woman using the nickname Xiao Ying created an AI version of her own mother and had chatted with her almost every day since last November. Because she is over eighteen, the app allowed her to keep her artificial mother. After reading the new rules herself, she expressed support for them and acknowledged that she understood the need for stricter oversight of AI companion use. Some apps have reportedly since added notices stating that messages were generated by artificial intelligence. However, she refused to name the three apps she uses out of fear that someone might shut them down.
Elsewhere, authorities are proceeding more cautiously than China
Beijing likes to present its tough approach as an effort to become a global leader in artificial intelligence safety. And it is not alone in its concerns; it is simply going further than others. The European Union has banned AI that encourages dangerous behavior among vulnerable groups, including children, and has required greater transparency from chatbots. In the United States, more than a dozen states have adopted their own rules for AI companions, primarily to protect minors, and a proposal for a nationwide ban for children is before the Senate. California and New York already have their own safeguards. China’s intervention, however, is stricter than any of these and is one of the few that does not give users the right to take their data elsewhere.
Sources: cnn.com, yahoo.com and theconversation.com
