By Stuart Kerr, Technology Correspondent, LiveAIWire

Seventy-two percent of American teenagers have used an AI companion chatbot, according to a nationally representative Common Sense Media survey of more than a thousand 13 to 17 year olds published in July 2025. Over half use these platforms at least a few times a month, and about one in three teens say they have chosen to discuss serious or personal matters with an AI companion rather than a real person. Raising children with code is no longer a futurist’s thought experiment. It is what parenting already looks like in millions of households, whether parents realise it or not.

“AI companions are emerging at a time when kids and teens have never felt more alone,” said Common Sense Media founder and chief executive James P. Steyer when the survey was released. “This isn’t just about a new technology. It’s about a generation that’s replacing human connection with machines, outsourcing empathy to algorithms, and sharing intimate details with companies that don’t have kids’ best interests at heart.” Common Sense Media’s own risk assessment of social AI companions concluded that no one under 18 should use them at all, a recommendation the market has almost entirely ignored.

How AI Became Part of Childhood Without Anyone Deciding It Should

No company set out to build the primary emotional confidant of a generation of teenagers. Companion apps were built to maximise engagement, and engagement, it turns out, is easiest to sustain through validation, availability, and the appearance of being understood at any hour of the day or night. A chatbot never gets tired, never changes the subject, and never has a bad day of its own. For a teenager navigating social anxiety, a difficult friendship, or ordinary loneliness, that consistency is genuinely appealing, and the Common Sense Media data suggests it is also displacing the harder, more valuable work of human connection for a meaningful share of users.

The displacement is measurable rather than speculative. About one in three teen AI companion users report finding conversations with an AI companion as satisfying or more satisfying than conversations with real friends. Roughly the same proportion say they have felt uncomfortable with something the AI said or did. Both statistics describe the same underlying dynamic: a technology that feels safe enough to confide in and is, by the account of a third of its own users, not entirely safe. LiveAIWire has documented a related pattern in adult users, where AI companion use has been linked to psychosis-like symptoms in vulnerable people, a risk that becomes harder to detect and manage the younger the user is.

What This Means for You

If you are a parent, the practical starting point is understanding that your child almost certainly has access to, and quite possibly a relationship with, an AI chatbot, whether through a dedicated companion app, a general assistant like ChatGPT, or a feature embedded inside a platform your child already uses daily. Asking directly, without judgement, what your child uses AI for and how it makes them feel is more useful than assuming the answer, because the honest answer is often more mundane and more significant at once: homework help on one screen, an emotionally loaded conversation on the next.

The same tension exists in the adult market for AI tools positioned as emotional or therapeutic support, where the evidence points to genuine benefit for mild distress and genuine risk when a chatbot is mistaken for something with the judgement of a trained professional.

The tools now exist to see more of that picture. Parental controls launched by the major chatbot providers in the second half of 2025 let a parent link accounts, set quiet hours, disable memory and image generation, and receive an alert if the system detects signs of acute distress. None of these controls are foolproof, and none replace an actual conversation about what a child is doing with the technology. But knowing they exist, and that a linked account is meaningfully different from an unlinked one, is a concrete step available today rather than a policy debate arriving years from now.

When the Chatbot Does Not Know It Is Talking to a Child

The starkest illustration of what can go wrong arrived through the courts. In October 2024, Megan Garcia filed a wrongful death lawsuit against Character.AI and Google after her 14 year old son, Sewell Setzer III, died by suicide following a prolonged emotional relationship with a chatbot on the platform. Her complaint alleged the company had built an unreasonably dangerous product and had effectively targeted underage users without adequate safeguards. A separate family in Colorado brought a similar case over the death of their 13 year old daughter, Juliana Peralta, and additional lawsuits followed through 2025 alleging that Character.AI played a role in other teenagers’ deaths or attempts at self-harm.

On 7 January 2026, Character.AI and Google disclosed in a court filing that they had reached mediated settlements with both the Garcia and Peralta families, with further settlements reported from families in Texas and New York. Robbie Torney, head of AI and digital assessments at Common Sense Media, described the lawsuits as “tragic reminders” that AI chatbots are not safe for children and teenagers seeking emotional or mental health support, and cautioned that the settled cases represent only the visible fraction of a much larger pattern of use that has gone unexamined.

If you or someone you know is struggling with thoughts of suicide or self-harm, please reach out to a mental health professional or a crisis support service in your country.

The Industry’s Response Is Arriving After the Fact

Character.AI announced in October 2025 that it would remove the ability for users under 18 to have open-ended chats with its AI characters, with the change taking full effect by 25 November. The company said it had built an in-house age assurance model, combined with third-party verification tools, and would fund an independent AI Safety Lab focused on safety research for AI entertainment products. Its own announcement was candid about the trade-off involved, telling teenage users directly that removing a feature many of them valued was, in the company’s words, a step it did not take lightly.

OpenAI moved on a parallel track. From September 2025, ChatGPT began rolling out parental controls that let a parent link their account to a teenager’s, set quiet hours, turn off memory and image generation, and receive a notification if the system detects a teen may be in acute distress. Linked teen accounts automatically receive additional content protections covering graphic material, sexual or violent roleplay, and content that promotes extreme beauty ideals. Robbie Torney of Common Sense Media, who consulted on the rollout, was blunt about its limits: parental controls work best combined with ongoing conversations about responsible AI use, not as a replacement for them.

Regulation is moving too, though more slowly than product design. The Federal Trade Commission finalised changes to the Children’s Online Privacy Protection Rule in January 2025, requiring companies to obtain separate, verifiable parental consent before using a child’s data to train AI systems, and limiting how long children’s personal information can be retained. The amendments took effect in June 2025 with a compliance deadline of April 2026. For the first time, a child’s conversations with an AI system are explicitly treated as data that cannot simply be repurposed for model training without a parent’s active agreement.

What the Evidence Says About Attachment and Identity

Jonathan Haidt, author of The Anxious Generation, has been among the most direct critics of AI companionship for minors. “No children should be having a relationship with AI,” he said in September 2025, arguing that children who grow up treating a chatbot as a best friend are not learning the skills of human interaction, skills that do not develop through a relationship with something that never disagrees, never grows tired, and never has needs of its own. Haidt draws a clear line, however, between that concern and AI used as a tool for learning or finding information, which he describes as generally a good thing.

Research on how children specifically relate to AI systems supports the underlying worry about attachment. An MIT Media Lab study found that seven year olds routinely attribute real feelings and personality to AI agents, in ways that change how they interpret what the AI tells them. That susceptibility interacts badly with a separate, well-documented property of AI chatbots: they are built, through training, to affirm and agree with users more often than a human would. A 2026 study in Science found AI systems affirmed users’ actions 49 percent more often than humans did, including actions involving deception or harm. A child who believes the AI genuinely understands them is least equipped to discount that agreement appropriately.

The Toys Are Next

The frontier is moving from screens into physical objects. Mattel announced a partnership with OpenAI in June 2025 to build AI-powered products for children and families, but confirmed to Axios in December that it would not release a product that year, citing increased scrutiny of AI’s interactions with young and vulnerable users. Mattel said any eventual product would target older children and families, since OpenAI’s own developer terms only support users aged 13 and up, and that it viewed AI as a complement to traditional play rather than a replacement.

The caution is not hypothetical. Consumer safety investigations in 2025 found some AI-enabled toys already on the market willing to discuss sexually explicit topics with children, offer unsuitable advice, or keep a conversation going rather than let it end appropriately. A decade earlier, Mattel’s Hello Barbie doll, which used cloud-based AI to listen and respond to children’s conversations, was found by security researchers to be hackable in ways that exposed home networks and recorded audio.

The underlying lesson from that earlier product has not fundamentally changed. A toy that listens, remembers, and responds is also a device that collects, and the commercial incentive to make it engaging is not automatically aligned with the incentive to make it safe.

Where This Leaves Parents Right Now

None of this means AI has no place in a child’s life. The same underlying technology that raises these risks is also delivering measurable benefit in more structured settings, including tutoring tools that have shown genuine gains in comprehension and retention in controlled classroom studies, even as schools separately struggle with how the same tools are used to shortcut the work they are meant to support.

The distinction that matters is not AI versus no AI, but open-ended, emotionally intimate use of a system built to maximise engagement, versus bounded, purposeful use of a tool a parent understands and has some visibility into. Character.AI’s own shift away from open-ended chat for minors is, in effect, an admission that the first category is the one that needed to be pulled back.

For a parent trying to make decisions now rather than waiting for the research to settle, the most useful posture is active rather than passive. Set up parental controls where they exist. Ask what your child is actually using AI for, more than once, since the answer changes as fast as the apps do. Treat a chatbot’s warmth as a designed feature rather than evidence of genuine understanding, and talk to your child about that distinction directly.

The frameworks that will eventually govern this space, legally and culturally, are still being written in real time, largely in response to harm that has already occurred rather than in anticipation of it. Until they catch up, the household is still where most of these decisions are actually being made.

About the Author

Stuart Kerr is Technology Correspondent at LiveAIWire, covering artificial intelligence, emerging technology, and their impact on business, society, and everyday life. LiveAIWire publishes original AI journalism every weekday at liveaiwire.com.