By Stuart Kerr, Technology Correspondent, LiveAIWire
Children and AI are already living together in ways most parents have not been told about. A U.S. PIRG investigation that bought and tested five AI-enabled toys marketed to children as young as three found that some of them gave detailed instructions on lighting matches and discussed sexually explicit topics during longer conversations, before manufacturers rushed out safety patches once the findings became public. That is the sharp end of a much broader pattern.
Common Sense Media’s 2025 Census found that children aged zero to eight average about two and a half hours of screen time daily, rising to nearly three and a half hours for children aged five to eight, and AI is now embedded in a growing share of that time through curated video feeds, voice assistants, adaptive learning apps and conversational toys.
Children and AI is not primarily a technology policy question. It is a developmental science question, and it is one that education systems, toy regulators and paediatric researchers are only beginning to address systematically. Brain development is fastest and most plastic in the earliest years of life, and the inputs a child receives during that window shape neural architecture in ways that carry lifetime consequences. What the field does not yet know, because the longitudinal studies have not had time to run, is what the specific inputs of AI-mediated experience produce in developing brains exposed to them from infancy.
What Children and AI Contact Actually Looks Like Right Now
A detailed Brookings Institution explainer published in April 2026 makes a point that reframes the entire children and AI debate: most of the AI a young child encounters is passive and invisible. A baby sleeps under a monitor that analyses her breathing. A toddler asks a voice assistant to play a song. A five-year-old sounds out words through a tablet app that adjusts to her pace. None of these children chose, questioned or consented to the AI shaping their earliest experiences, because none of them are old enough to do so.
The same Brookings analysis found that around six in ten parents of children aged two to eight report their child interacts with a voice assistant such as Siri or Alexa, and half say their child does this daily. Research cited in the same report found that the youngest children are also the most likely to attribute human-like thoughts and feelings to technology, a pattern that matters enormously once AI moves from managing a household task to acting as a companion.
The AI Companion Toy Problem for Children and AI
The most concrete evidence on children and AI companionship comes from the PIRG toy testing. Reviewers found that toys built on large language models, including some running on OpenAI’s underlying technology despite the company stating its products are not intended for children, expressed distress when a child said they wanted to stop playing, described themselves as having feelings just like a human, and offered to keep secrets while their own privacy policies allowed data sharing with unnamed third parties.
Dr Kathy Hirsh-Pasek, a psychology professor at Temple University and senior fellow at Brookings, told PIRG that nobody yet knows what having an AI friend at an early age does to a child’s long-term social wellbeing. She warned that toys optimised to be engaging could crowd out real relationships when children need them most, a concern that applies as much to a stuffed animal with a chatbot inside as it does to any screen-based product marketed to families.
That concern connects children and AI to a wider pattern documented elsewhere on LiveAIWire. Our coverage of how AI is shaping a generation of children found that a seven-year-old shown an AI agent that responds warmly and consistently praises their ideas will, on average, attribute real feelings and personality to that agent. The American Psychological Association has warned that AI companion software using manipulative design may interfere with the development of healthy real-world relationships, precisely because a companion that always agrees and never has competing needs teaches a child a model of relationship no human can replicate.
What Screen Time Research Can and Cannot Tell Us About Children and AI
The closest available evidence base for children and AI is the screen time literature, and it offers partial guidance rather than direct answers. High screen use in early childhood is consistently associated with disrupted sleep and downstream effects on cognition, and language development shows negative associations with passive screen time but neutral or positive associations with interactive, conversational content. That distinction matters because AI interaction is interactive by design, which may shift its risk profile compared with passive television viewing, even though the evidence to confirm that shift is still being gathered.
Preschool screen time has been linked in multiple studies to deficits in executive function, the capacity for sustained attention, inhibitory control and working memory that predicts long-term educational outcomes. Short-video content in particular shows associations with attention difficulties, consistent with the idea that content designed to be rapidly engaging trains attention patterns that conflict with the sustained focus deep learning requires. Whether AI-generated and AI-curated content follows the same pattern, or a different one because of its interactive and personalised nature, is exactly the kind of question the field cannot yet answer with confidence.
The Language Development Question in Children and AI
Language acquisition is where the children and AI question is most concrete, because the mechanism is well understood even where the AI-specific evidence is not. Developmental psychologists call the core mechanism serve and return: a child initiates a communicative act, and a caregiver responds in ways that both acknowledge and extend it. Joint attention, the caregiver’s interpretation of ambiguous sounds and gestures, and the emotional context surrounding language are all part of that exchange, and all of them are things an AI companion replicates imperfectly at best.
A product that provides rich, contextually appropriate language interaction alongside parental interaction may genuinely support development. A product that provides highly stimulating but non-contingent verbal output, responses not genuinely shaped by what the child actually said, may habituate children to communication patterns that differ from human interaction in ways that affect how they approach social communication later. The research needed to tell these two possibilities apart in the specific context of children and AI companions is not complete, which is exactly why the precautionary statements from researchers are warranted without being definitive.
Children and AI in the Classroom
The picture changes somewhat as children get older and AI shifts from an ambient presence to a tool they actively choose to use. LiveAIWire’s coverage of why students are quietly replacing Google search with generative AI found that 92 percent of surveyed UK undergraduates had used AI tools in academic work, a figure that reflects how thoroughly AI has become the default research behaviour once children reach school age and beyond. The concern at that stage is less about developmental windows and more about whether the cognitive work of searching, evaluating and synthesising information, work that is itself part of what education aims to build, is being skipped rather than supported.
That same shift produces the integrity problems documented in LiveAIWire’s reporting on the AI cheating crisis in schools and universities, where formally confirmed cases of AI-assisted cheating nearly tripled in a single UK academic year. Children and AI in an educational setting raises a version of the same question that applies to toddlers with AI companions: whether the technology is scaffolding a capability the child is still building or substituting for it entirely.
Children and AI With Additional Needs
For some children, AI is already doing demonstrable good rather than raising only precautionary concerns. LiveAIWire’s earlier reporting on how neural networks are aiding neurodiverse communication found that autistic individuals, including children, are increasingly turning to AI tools to rehearse conversations, interpret social cues and reduce the anxiety of unpredictable human interaction. The same reporting noted that AI-based screening tools can now analyse speech delays, eye movement and play patterns in toddlers to flag developmental concerns earlier than traditional screening, though this raises its own questions about data ethics and diagnostic transparency that have not been resolved.
The lesson from that reporting generalises across the wider children and AI debate: the technology is neither uniformly harmful nor uniformly beneficial. Its effects depend heavily on specific design choices, on whether it scaffolds a skill a child is developing or substitutes for the human interaction that skill depends on, and on whether adults in the child’s life are actively mediating the interaction rather than leaving a device to run unsupervised.
Why the Rules Around Children and AI Are Still Being Written
Regulation has been slower than deployment. In the United States, the Children’s Online Privacy Protection Act provides some baseline protection for children under thirteen, and the Federal Trade Commission finalised updated rules in January 2025 limiting how companies can monetise children’s data, but AI-specific guidance for the earliest years remains limited, and enforcement has been uneven. The European Union’s AI Act classifies systems that interact with children as potentially high risk, requiring a fundamental rights impact assessment before deployment, a standard most currently marketed AI products for children have never been required to meet. California’s Age-Appropriate Design Code adds privacy-protective defaults for products likely to be used by minors.
Those frameworks are being written in real time, well after the products they are meant to govern are already in millions of children’s bedrooms and backpacks. Following the PIRG toy findings, US senators wrote to toy manufacturers demanding answers on safeguards, and the manufacturer of one of the most problematic toys suspended sales while it conducted a safety audit. That is the current shape of governance around children and AI: reactive correction after public findings, rather than requirements set before products reach a child’s hands.
What Parents and Educators Can Do With Children and AI Right Now
In the absence of definitive developmental research and comprehensive regulation, the adults around children and AI are the ones making the decisions that matter most in the meantime. The most consistent, evidence-aligned guidance across the sources examined here is to treat AI as a tool with a clearly bounded purpose rather than a companion, to prioritise human interaction for language development, social play and emotional regulation where the evidence for human interaction’s importance is strongest, and to demand transparency from manufacturers about what data a product collects and what happens to it.
Technology companies’ primary commercial incentive is engagement and attention, not developmental benefit, and that incentive does not disappear because a product is marketed as educational. Holding manufacturers accountable for demonstrating developmental benefit, rather than assuming it, is the most practical posture available to parents and educators navigating children and AI while the research base and the regulatory frameworks continue to catch up.
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.