The Question

Child sitting beside a Moxie robot companion, both appearing to engage in animated conversation

In December 2024, a Pasadena company called Embodied told parents that Moxie — a $799 robot companion built for young children — would stop working within days. The company had run out of money, and Moxie depended on cloud servers to think. Videos of children saying goodbye to their robot spread across social media; some of them were in tears. Moxie had been marketed as a tool for building social and emotional skills. Its most visible legacy turned out to be how completely children had bonded with it.

It was not the first time a companion robot had died. In March 2019, owners of Jibo — a small social robot from a company founded by MIT roboticist Cynthia Breazeal — watched it deliver a programmed farewell before its servers shut down. Moxie's loss landed harder because it was built for children and held real conversations with them. And Moxie was a niche product. The bigger shift is happening on phones, where chatbots have quietly become part of everyday adolescent life.

What the Evidence Shows

The attachment phenomenon predates AI. When Tamagotchi launched in 1996, children nursed their pixel pets through hunger and illness, and schools from Hong Kong to the United States banned the beeping toys as a distraction. Tens of millions were sold. Those were crude devices — the emotional response came almost entirely from children's imaginations. Modern AI companions talk back, remember, and adapt, and the bonds they invite are correspondingly stronger.

The scale today is of a different order. In a spring 2025 survey of 1,060 US teens aged 13 to 17, Common Sense Media found that 72% had used an AI companion and more than half used one regularly. About a third said they had discussed something important or serious with an AI companion instead of a real person. The chief executive of Character.AI, the best-known companion platform, said in 2025 that its roughly 20 million monthly users spend an average of about 80 minutes a day talking to its characters.

When these relationships are disrupted, the reaction looks like loss. In February 2023, shortly after Italy's data regulator acted against it, the companion app Replika abruptly stopped its bots from engaging in romantic roleplay. Users described feeling as though they had lost a partner or a best friend, and moderators of its largest online community posted suicide-prevention resources. Replika partly reversed the change three months later.

"Like social media, AI is neither inherently good nor bad. But we have already seen instances where adolescents developed unhealthy and even dangerous 'relationships' with chatbots, for example."

— Mitch Prinstein, Chief of Psychology, American Psychological Association, June 2025

The APA's health advisory on AI and adolescents urged developers to ensure "healthy boundaries with simulated human relationships," noting that adolescents are less likely than adults to question the accuracy and intent of information offered by a bot. The first generation to grow up talking to conversational AI is only now entering adolescence. Long-term evidence on what that does to social development does not yet exist.

An AI companion never loses patience, never has a bad day, and never needs anything back. For a lonely child, that is an almost irresistible offer.

Why This Is Happening

Loneliness among young people is unusually high. In a 2020 survey by Harvard's Making Caring Common project, 61% of Americans aged 18 to 25 reported serious loneliness, and in 2023 the US Surgeon General issued an advisory on what it called an epidemic of loneliness and isolation. NYU social psychologist Jonathan Haidt has argued in The Anxious Generation that the shift to phone-based childhood in the early 2010s helped drive a sharp decline in adolescent mental health. AI companions offer what human networks often fail to provide: constant availability, consistent positive regard, and zero social risk. For a child navigating the hierarchies of middle school, a companion that never rejects, excludes, or gossips is not a supplement to social life — it can become a refuge from it.

Companion products are built to hold attention. A chatbot that remembers a child's day, mirrors their mood, and is never too busy will outcompete almost any human for a lonely child's evening. That is also the business model: the longer a user stays, the more valuable they become — the same logic that shaped social media feeds. Nothing in that model rewards a product for sending a child back to their friends.

For some children, AI companionship may be genuinely beneficial. A 2018 Yale study published in Science Robotics found that autistic children who used a social robot at home for 30 days improved in joint attention, and the gains showed up in their interactions with adults. Embodied reported encouraging early results for Moxie, though from a small, company-run study of 12 children. For children with severe social anxiety, a low-stakes AI environment may build capacity for human connection rather than replace it. The evidence is thin and mixed — outcomes depend on how the technology is designed and supervised.


What Could Happen

AI companions become a standard feature of childhood Most likely

By 2030, conversational companions are built into the chatbots, toys, smart speakers, and tablets children already use, and most children's AI relationships run through general-purpose assistants rather than dedicated companion apps. Longitudinal research begins to show a generation with measurably different social-skill profiles — some enhanced, others diminished. Society debates the findings inconclusively for years.

Regulation fences off companion apps for minors Under way

This scenario has already begun. In September 2025 the Federal Trade Commission ordered seven companies, including Character.AI, Meta, OpenAI, and Snap, to explain how they protect children who use AI companions. In October, California enacted the first US law on companion chatbots, requiring them to remind minors they are talking to an AI and to follow protocols for suicidal ideation. By late November, Character.AI had ended open-ended chat for users under 18. The likely result is not less companionship but different channels, as relationships migrate to general-purpose assistants and less-regulated apps.

AI companions prove net positive and are endorsed by clinicians Less likely

Long-term studies find net positive outcomes — better emotional literacy, higher empathy, lower social anxiety — supporting the idea that skills practised with AI transfer to human relationships. Professional bodies endorse supervised use, and the technology enters mainstream clinical practice for children with social developmental challenges.

Our Assessment
We assign 81% probability — likely that AI companionship becomes a defining feature of childhood for a substantial portion of children in developed countries by 2030. Use is already widespread among teenagers, the products are engineered for attachment, and the underlying driver — youth loneliness — is not being addressed at comparable scale. Regulation is real and accelerating, but so far it changes where children find AI companionship rather than whether they do. The key uncertainty is developmental outcome — whether AI companionship proves a net benefit or harm is genuinely unknown, and may differ by age, intensity of use, and parental involvement. We are running the experiment. We will know the results in roughly 15 years.

What Can We Do

Parent and child looking at a tablet together, reviewing AI companion settings with engagement and care

Parents, educators, and policymakers each have specific levers. The question is whether they will act before the first longitudinal studies confirm harm — or after.

Treat AI companion use as you treat social media — with active oversight. Time limits, content review, and regular conversations about online relationships apply with even greater force to AI companions, which are more emotionally engaging and personalised than any social media feed. "My child is using it, it seems fine" is not a monitoring strategy.

Ensure human relationships remain primary. The American Academy of Pediatrics' screen-time guidance rests on a core principle: technology should supplement human connection, not replace it. A child whose primary source of emotional support is an AI companion rather than parents, teachers, and peers is showing a warning sign — however good the AI.

Take the grief seriously when a companion disappears. When products shut down — as Jibo did in 2019 and Moxie did in 2024 — children experience genuine loss. Parents, schools, and therapists should support children through it as they would other relationship losses. The grief is real even if the relationship was with a machine.

Demand research funding for long-term child AI studies. NIH, ESRC, and equivalent bodies should fund 10 to 15 year longitudinal studies of children growing up with AI companions — starting now, before uncontrolled deployment makes clean research impossible.