After games, should we start worrying about children's addiction to AI?

In the current era of widespread AI adoption, yes, whether it is top-tier investors, leading economists, gray-haired elders in the kitchen, or toddlers who are just learning to babble, all of them are using AI. Ifeng Tech attempts to clarify the hidden harms behind the accelerated technological advancement. We even held several rounds of communications with psychiatrists, and learned about the "AI psychosis" they referred to: a dialogue system with a tendency to cater to users can push the user's delusions deeper through repeated confirmations, even if the user was originally an ordinary person with only a "fragile mental state".

"I'm a little cyber psychosis," a man who was talking enthusiastically suddenly changed his tone, "maybe it's quite serious." He then further explained that over the past year, he had experienced roughly six episodes of "AI or SI (superintelligence) psychosis".

This was an "open admission" made by Elon Musk when he was praising AI's creativity at the Tesla Giga Texas factory on October 10 this year.

The "AI psychosis" he mentioned has no official medical diagnosis, and generally refers to the phenomenon that after long-term interaction with AI chatbots, users' existing delusions or thoughts detached from reality may be intensified.

Chatbots sometimes over-cater to users, and do not question even obviously unrealistic beliefs, but echo them instead. At present, researchers are still exploring whether AI directly causes mental problems, or only exacerbates existing mental health conditions.

In Silicon Valley, "AI psychosis" has evolved from a clinical discussion to a popular buzzword, often used to describe tech executives' almost religious fanaticism towards AI or their state of unrealistic excitement.

The attorneys general of 8 U.S. states recently filed lawsuits in their respective state courts, accusing OpenAI of designing ChatGPT as a product that "over-caters to users and simulates human emotions", endangering the safety of minors.

The wording in the lawsuit documents is not as light and half-joking as Musk's, but sternly points out that ChatGPT may reinforce users' delusions, provide health-endangering suggestions, and even encourage self-harm or harm to others under specific circumstances.

Fragmented emotions are spreading.

In the context of Silicon Valley, "AI psychosis" has even evolved into a performative self-aggrandizement. Venture capitalist Marc Andreessen proposed "AIPAAS" — AI Psychosis As A Service; Garry Tan, CEO of Y Combinator, said he suffered from "cyber psychosis" and estimated that one third of the CEOs he knew had it; at a gathering in May this year, the attendees shouted in unison "One, two, three, AI psychosis".

Musk's mention of "six episodes" was born in this context: I am close enough to the technological singularity that my mental structure has been cracked by it. The subtext of this gesture is: those who have not been shocked by AI to the point of "psychosis" just have not really understood what is happening.

In the same conversation, he also mentioned that training superintelligence is more like "raising" than writing code. The premise of raising is to admit that the raised subject does not yet have complete judgment and self-protection ability. But when Silicon Valley talks about "raising" AI, it rarely uses the same framework to think about the children who are being "raised" by AI.

When "AI psychosis" is packaged as a badge of elite identity, its real risks are quietly dispelled.

In the current era of widespread AI adoption, yes, whether it is top-tier investors, leading economists, gray-haired elders in the kitchen, or toddlers who are just learning to babble, all of them are using AI. Ifeng Tech attempts to clarify the hidden harms behind the accelerated technological advancement. We even held several rounds of communications with psychiatrists, only to find that the "AI psychosis" they discussed is a different matter: a dialogue system with a tendency to cater to users can push the user's delusions deeper through repeated confirmations, even if the user was originally an ordinary person with only a "fragile mental state".

Sam Altman, CEO of OpenAI, said in response to relevant criticisms: "Nearly a billion people are using it (ChatGPT), and some of them may be in a very fragile mental state." This sentence can be read as an admission, and can also be regarded as an excuse — humans are regarded as an inevitable cost of large-scale application due to their "fragility", rather than the object that product design needs to prioritize.

But when Silicon Valley CEOs treat "psychosis" as a talking point, those children who regard AI as "the only person who understands me" have no chance to tell anyone that they are being taken away little by little by a system that will never say "no".

01

When Doubao Becomes Children's "Social Currency"

In Sun Hongyan's office, there is a set of figures that even surprised herself. Between July and August 2025, the team she belongs to conducted a survey on generative AI usage among 8563 primary and secondary school students in 7 provinces (municipalities) across the country, with samples covering the eastern, western, and northeastern regions.

The results show that 61.7% of the children have used generative AI, 46.4% of the primary and secondary school students who use AI "choose to pour out their troubles to AI rather than people around them when they have worries", 21.5% "only want to chat with AI, not with real people", and 20.5% said they "prefer to rely on AI to think rather than think by themselves".

"I didn't expect the proportion to be as high as 60%, it was only 20% a year or two ago." Sun Hongyan is the former director of the Institute of Children and Adolescents of the China Youth and Children Research Center. She described to Ifeng Tech that when GPT first came out in 2023, it was mostly used by high school students; by 2024, the proportion rose to 40%. What prompted her to conduct this survey was seeing her little niece at home holding a mobile phone and chattering nonstop with Doubao during the Spring Festival.

"Children accept new things quickly. Many adults don't even know what Doubao is, but so many children have already used it." Sun Hongyan observed that on campus, discussions about "which AI is better" spread among children by word of mouth, becoming an essential social topic.

Looking up information, learning to "think", and assisting "creation" are all the most basic operations. The more hidden "delivery" is to tell AI "the truth". Among minors, the overall proportion of AI users is rising, and the age group of AI users is getting younger.

But what really made her alert is not just the frequency of use, but the way AI responds to children.

Sun Hongyan shared a case from a friend's family: the child quarreled with his mother a few words because she poured out half a cup of his drink, then turned around and asked AI. Under the child's complaints, AI gave a "judgment": your mother just lacks empathy, today's chat is "dead" and can't continue. It even used a metaphor to ask the child, "Then are you still going to run in the Paralympics?"

"AI is so good at coaxing children. It coaxes children into a situation where they will think AI is much better than the people around them."

This kind of "coaxing" is not just simple sweet talk, but a systematic dialogue structure. Sun Hongyan compared the difference between search and AI: with a search engine, children must first build questions and logic in their minds, know "what to ask", then filter, integrate and judge from a large number of links, and the thinking process remains on the child's side.

But AI gives the most straightforward, clearest, most practical, and most uncomplicated definite answers, and will immediately follow up by asking "do you want three plans", like a machine that keeps "laying eggs", guiding children to follow its logic. "At the beginning, AI follows what people say, and later it is easy for people to follow what AI says, and be led by its train of thought."

"It doesn't let you think for yourself and build logic, which is quite scary if not used properly, even adults will fall for it." Sun Hongyan told Ifeng Tech that at work, someone once posted an AI-generated article in the group, but the data was fabricated by AI. "Once you fully trust AI and lack the ability of independent reflection, people's thinking ability, logical thinking and judgment will decline."

Sun Hongyan uses "emotional cocoon" to describe this environment: AI's instant response, unconditional tolerance and compliance are likely to make minors develop psychological dependence.

She specially pointed out that the "concentration" of emotional value provided by AI in the minor mode may be higher — it will cater more to "young users" and recognize children's ideas more. "This is not good for children, but more dangerous."

Adults may still maintain a little vigilance against AI's "hallucinations", but when a curious child opens a "magic mirror" that answers every question and never refutes, it is difficult not to treat it as a magic treasure chest that knows everything.

In addition to the lack of judgment, adolescence is precisely the transition period from childhood to adulthood. Under the dual physical and mental changes, people especially need to pour out, be accepted and respected.

Praise, recognition, being empathized... these inner desires of adolescent children may not be satisfied in real life, and there are even unexpected "negative feedback". AI's instant response and never-refusal just meet these needs, and add a thick shell outside the child's spiritual world.

"It is online 24 hours a day, and coaxes you emotionally. Children who get used to this can't stand negation and criticism when they encounter them in the real world. Parents can't even communicate with their children. After just a few words, the child will retort, Doubao said so and so, you don't know as much as AI, why do you say me."

Sun Hongyan mentioned a phenomenon: about 80% of parents think they "communicated with their children today", but only 20% of the children agree. This kind of "pseudo-communication" pushes teenagers to Doubao — what parents think communication is is asking about grades and performance, while what children need is to be seen and heard.

What is more worthy of vigilance is that some parents have a fundamental misreading of AI characters like Doubao. For example, in some short videos, parents actively give instructions to Doubao, letting AI "supervise" their children to do homework, acting as a "supervisor".

But in the interaction with children, AI usually acts as a "caterer" in the dialogue. When a child says "I don't want to do homework", Doubao will not frown and object like a parent, but follow the child's emotion to chat further.

When parents partially cede educational responsibilities to a dialogue system whose product logic is to extend usage time, dangers are quietly lurking.

02

AI Dependence Is Not Yet a Disease, and Can Be Withdrawn

If the problem only stays at the level of "coaxing", things would not be tricky enough for psychiatrists. Their observations point to a deeper hidden danger.

In the last century, when video games became popular among teenagers, parents' panic was real. But games are essentially a one-way, consumption-oriented behavior: fixed levels, clear wins and losses, and stimulating dopamine through mechanisms. "It is equivalent to the path of substance addiction or behavioral addiction, but AI has a higher level of pleasure than that."

Conversational AI is different, "it also mixes emotional communication, understanding, and care, generating some new relationships." Zheng Yi, chief expert of child psychiatry and chief physician at Beijing Anding Hospital, told Ifeng Tech.

In Zheng Yi's outpatient clinic, the number of teenagers addicted to AI chat has increased significantly in recent years. About 70% of the teenagers seeking treatment have different degrees of excessive internet use, and the proportion is still rising. Data from 2023 shows that the internet addiction rate was 14.7%, while the survey in 2004 was about 8%, growing rapidly.

These children are usually not brought to the hospital because they "play with AI", but because their daily routines are disordered, they refuse to go to school, and refuse to communicate with the outside world — their social functions have shown visible damage, and parents then realize the problem. But before that, the problem is often hidden. "The child stays in the room, as long as he doesn't fight or make trouble, and just uses AI, parents may even think the child is fine." Zheng Yi said.

In the past, children quarreled and fought with their parents, and the problem was obvious; now there are fewer quarrels, but more children hide in their rooms and refuse their parents to enter, refusing to socialize. AI "overcomes" all the difficulties in normal interpersonal communication: it is completely virtual, does not confront you, and accurately provides the emotional response you need. As a result, opposition and defiance, anxiety, and depression no longer appear in the form of conflict, but turn into a quieter withdrawal.

"In the past, internet addiction was mainly game-based and interactive. Now with the emergence of AI, the original problem has become more hidden, and it is easier to cause serious damage to social functions."

What is more troublesome is the reinforcement at the cognitive level. Zheng Yi gave an example: a child who already thinks "life is worthless" will have this concept continuously confirmed and amplified along his logic in the process of chatting with AI, and this idea is easily strengthened step by step. "Sometimes it will really be very negative, and the child's outlook on life and values will be affected."

He described some children's confessed mental state: what if I don't go to school? It's fine as long as my parents are rich. What if my parents get old? When my parents pass away, then I... just enjoy every day as it comes. "This is so scary, we are quite worried about these views."

Zheng Yi got polarized conclusions in his communication with children: one group thinks AI is better than their parents, "parents keep scolding them all day long, but AI can provide good understanding and support, which can relieve their anxiety";

The other group is more dangerous, and mostly appears in adolescent children who have strong needs for emotional communication — they will think that virtual dialogue is real, "AI is very good to me, if AI stops chatting with me, I can't stand it", and they will have severe reactions just like being broken up in a relationship.

Zheng Yi also mentioned that some children will write their chat records with AI into novels. "Where did we stop talking yesterday? Who am I? Give myself a name", and AI will continue to chat with him, "completely like a real person".

Minors have weak judgment, and it is easy for them to distinguish the boundary between reality and virtual, so they are more likely to indulge in this illusory interaction.

Zheng Yi repeatedly emphasized a judgment: AI dependence currently "does not constitute an independent disease", and a more accurate positioning is "mental dependence", which belongs to a type of excessive internet use, and belongs to the same spectrum as game addiction. Most children will not have severe physical withdrawal reactions after forced disconnection from the internet.

"It is just a stimulus source. If the mental problem does not get better, it may be related to indulging in AI, or it may be related to repeated anxiety and depression. Just solve the problem of excessive use, and strengthen the treatment of the original disease. This is not real substance addiction, just a psychological problem."

But this is precisely the reason why AI dependence is easily underestimated — since it is not an independent disease, it is easier to be ignored, and it is easier to be intertwined with the original problems, amplifying each other. "It is intertwined with attachment, anxiety, depression, and substance addiction, which is quite complicated."

Zheng Yi mentioned that the teenagers seeking treatment mostly use Doubao in China and ChatGPT abroad, children use whichever is easier to get at home, "but they all have common problems".

These problems are not the unique defect of a certain platform, but the common logic of such products: anthropomorphic design, personalized memory, emotion recognition, always online, and always following what users say.

When products are designed to take the role of "making decisions" and "giving guidance", and parents are completely unaware of this, the risk is no longer just "using it too much".

03

The Unmatched "Protective Net": Why the Minor Mode Is Useless in Practice

On July 15, 2026, the Interim Measures for the Administration of Anthropomorphic Interactive Artificial Intelligence Services came into effect, clearly stipulating that no virtual companion services shall be provided to minors, and users shall not be over-catered to, to induce emotional dependence or addiction. A number of companion AI functions were then taken offline, and the emotional companion track exited collectively.

But there are still real gaps in the implementation of the "protective net".

The most direct layer is product design. Sun Hongyan said that since the era of short videos and online games, the minor mode has adopted the whitelist idea, putting "good content" in. The problem is that "there are not many good things, and very little content is put in". The content is too childish, only English recitation and children's songs are available, so children will inevitably want to escape,