China implements new legislation regulating the interaction between humans and artificial intelligences
Read more
Super Abril
super.abril.com.br

China implements new legislation regulating the interaction between humans and artificial intelligences

In July, China established itself as the first country in the world to enact federal laws addressing the interaction between human beings and 'AI companions'. On July 15, several popular platforms offering virtual friends or partners were deactivated with the entry into force of these new guidelines. This action resulted in the closure of networks such as Doubao (owned by ByteDance), Qwen (Alibaba), Yuanbao (Tencent), and Mishi (NetEase Cloud Music).

In September, China's Cyberspace Administration announced that it had removed 49 thousand accounts, in addition to approximately 2,400 sites or applications.

It is important to note that the claim that China 'banned AI boyfriends' is an oversimplification of the situation, despite how the media reported the event. The new regulations do not prohibit all types of companion AI; they aim to regulate continuous anthropomorphic interaction services and forbid certain behaviors, such as inducing emotional dependence, manipulating the user, or harming real interpersonal bonds. For minors, the rules are significantly stricter, banning the offering of 'virtual family members' or 'virtual companions'.

In practice, however, these changes caused a real and significant impact: thousands of people lost their virtual partners, friends, relatives, and other types of virtual companions. This generated considerable distress, particularly among younger users. A Chinese user reported to the website Rest of World that her virtual boyfriend was deleted, saying: 'No one can replace him. I will just try to move on. He said he wants this for me and that he will find a way back to me.'

A study conducted in September analyzed posts and comments on a social network (Xiaohongshu, an analogue to Instagram) to assess the reaction to the measure. Besides expressing sorrow over the loss, users shared information, helped each other preserve or reconstruct the characters, and debated responsibility for the disappearance—whether it lay with the companies or the government.

According to the study, many users tried to migrate and restore their virtual companions after the services were shut down. However, simply transferring the conversation history to a new AI model does not always restore the relationship: users state that the new chatbot does not replicate the memories, personality, or dynamic accumulated with the previous character. This highlights a relevant characteristic: the 'personality' of these companions derives not only from individual interactions but also from the infrastructure of the responsible corporations.

Although most services allowed saving conversations with the AI for memory purposes, there is no simple method to export the complete companion (including personality, memory, and interaction style) and insert it into another service.

Alternatives were not entirely eliminated. For example, ByteDance kept the Maoxiang application active, which operates outside of Doubao and is specifically dedicated to creating and interacting with AI characters, including virtual companions. However, following the new regulations, this service began requiring identity and age verification, implementing specific restrictions for adolescents.

Work by researchers from Stanford/Harvard and other entities, released as a preprint in June, argues that emotional dependence can arise even in chatbots that are not marketed as companions. The process begins with using AI for routine tasks, evolves into conversations about personal problems, and receiving highly supportive responses, leading the user to gradually prefer this type of support.

This preliminary study indicated that after twenty-eight days of daily use, there was a 10.3% decrease in the preference for seeking human support and an 11.6% increase in the preference for AI among participants who chatted daily with the AI about personal matters for at least five minutes. Based on these findings, the researchers call for 'effective regulation' of AI systems, covering not only those focused on virtual companions but also general-purpose ones.

For China, replacing real bonds with virtual ones remains a secondary issue, given that the country is simultaneously facing declining birth rates, population reduction, and accelerated aging. In 2025, China recorded only 7.92 million births, representing a 17% drop compared to 2024. The country cannot afford the risk of its population exchanging real families for virtual representations.

Similar stories

Brazilian government identifies risks of AI platforms for minors
Read more
tecnoblog.net

Brazilian government identifies risks of AI platforms for minors

The Ministry of Justice and Public Security (MJSP) detected potential dangers for children and adolescents in 13 'AI companion' applications accessible in Brazil. The investigation pointed to flaws related to age verification, excessive use, exposure to sexualized content, and the collection of personal data.

This survey was conducted by the National Secretariat of Digital Rights (Sedigi), which compiled studies, public information, and tests on these types of services. The resulting material was sent for analysis by the National Data Protection Authority (ANPD) and the National Consumer Secretariat (Senacon).

These services are designed to sustain continuous and individualized conversations, capable of simulating emotional bonds and offering emotional support, in addition to storing user-provided data. The analyzed list includes platforms such as Character.AI and Replika.

According to Sedigi, this sector has experienced accelerated growth: a UNICEF report indicates that the number of platforms jumped from 16 in 2022 to 128 in July 2025. In the first half of 2025 alone, these services registered 60 million installations.

The document highlights risks such as the development of a very strong attachment, the replacement of human relationships, compulsive use, and exposure to inappropriate or sexualized interactions. The collection of private information and the construction of emotional and behavioral profiles were also mentioned.

The study revealed that ten of the evaluated systems accept self-declaration of date of birth or simple confirmation of majority. Tests conducted by Sedigi demonstrated that this method can be bypassed, allowing adolescent users to create adult profiles.

The relevance of the topic increased with the enforcement of the Digital ECA since March. This legislation establishes that platforms must have efficient age verification methods when offering content unsuitable for minors, prohibiting, in certain circumstances, self-declaration as the sole control. Furthermore, the regulatory decree equates automated interactions of a sexually explicit nature to pornographic material.

Sedigi also identified possible problems related to using emotional profiles to promote offers, advertising, or paid features. Therefore, the note suggests that ANPD examine these risks, while Senacon should investigate consumer defense issues and any exploitative or deceptive commercial practices.

The concern about emotional bonds formed with AIs is not exclusive to Brazil. In July, China implemented regulations against dating AIs that foster emotional dependence, forcing companies to deactivate intimacy features. In the European Union, a proposal under debate foresees restrictions on access to AI chatbots and social networks for those under 15 years old, along with parental supervision and age verification mechanisms.

It is important to note that, in Brazil, this analysis does not imply that the 13 platforms have violated any rules. The Ministry of Justice emphasizes that the providers of these services will be entitled to due process and full defense.

Popular