Experts have warned that the rapid adoption of AI in children poses risks to cognitive development, data privacy and safety due to the lack of a pediatric framework. This has implications for education, edtech and digital platforms, increasing pressure on regulators and companies to adopt child-centric AI standards and strengthen data governance.
The United Nations has established the first independent international scientific panel on AI. This panel aims to address the rapid integration of these technologies, where one-third of the world’s internet users are children.
UN Secretary-General António Guterres explains that AI is moving at the speed of light and requires urgent scientific oversight. “Children are often not expected to be the users of digital resources, so their needs and rights are often not supported,” explains Sonia Livingstone, professor of social psychology at the London School of Economics and Political Science and member of the United Nations Committee.
AI is being integrated into search engines, social media platforms and educational tools without a specific framework for child safety and cognitive development, experts warn. Data shows that children are actively participating in generative AI environments.
According to a report by the EU Kids Online network, 72% of children aged 9 to 17 in the European Union are generative AI users. Adoption rates vary widely geographically, reaching 94% in Austria, 89% in Italy, and 88% in Serbia. Conversely, countries like Ireland and Spain have lower engagement rates at 40% and 47% respectively.
While AI-assisted learning offers potential benefits for students with disabilities, Livingston notes that a lack of long-term research on mental health, sleep cycles, and critical thinking creates a high-stakes environment. Mr Livingstone said AI tools were “harvesting” data without children’s consent or consideration of their developmental status.
Gap in parental awareness and reality of use
It has been documented that a disconnect exists between parental perceptions of AI and actual adolescent behavior. A study conducted by the Pew Research Center and Common Sense Media revealed a significant communication gap. Pew Research Center Managing Director Monica Anderson said 64% of teens ages 13 to 17 report using chatbots, but only 51% of parents know how to use them. Furthermore, 4 in 10 parents report that they have never discussed AI with their children.
This void is especially evident when using AI for emotional support. According to Pew Research Center, 12% of teens use AI for advice and companionship, and 16% have casual conversations with models. Demographic analyzes show significant racial disparities in these behaviors. 21% of black teens use AI for emotional support, compared to 13% of Hispanic teens and 8% of white teens.
Michael Robb, head of research at Common Sense Media, said a significant minority of children were using AI in social ways, which could make parents uncomfortable. These children may describe AI as their “best friend” or primary confidant, which is recognized as a red flag for questionable usage.
Cognitive withholding and cognitive atrophy
For adults with established expertise, delegating tasks to AI results in “cognitive atrophy,” or weakening of existing skills that can be recovered, says Psychology Today. The risk for children is “cognitive foreclosure.” When children delegate tasks they have not yet learned to perform, such as constructing arguments or evaluating sources of information, they avoid forming essential neural pathways.
A 2026 study by Cornell University researchers Judy Hanwen Shen and Alex Tamkin demonstrated this effect among software developers. Developers who fully delegated tasks to AI produced working code but failed subsequent conceptual quizzes. Their performance was 17% worse than the group that received no AI assistance. For children, this effect is even worse. The replacement of learning with AI will be permanent, as children lack the expertise to “audit” the output of AI.
Homogenization of reasoning and identity
Large-scale language models (LLMs) work primarily on the basis of statistical probabilities derived from Western-educated mainstream training data. When children process information consistently through these models, they run the risk of adopting the model’s inferential structure as their own. This presents a threat vector to the developing mind: homogenization.
The statistical biases of the model become the default framework for students. A study published in Cell Press claims that LLM homogenizes language, perspective, and reasoning strategies. For children who have not yet formed independent reasoning, this common output poses major identity problems. The model does not compete with the child’s reasoning. replace it.
Rapid increase in AI-generated content
Automated production has visibly reduced the quality of content in the pediatric digital space. According to a Kapwing report published in November 2025, approximately 21% of YouTube feeds consist of low-quality AI-generated content, often described as “sloppy.” These videos are produced on an industrial scale.
Dana Suskind, a professor of surgery and pediatrics at the University of Chicago, calls this a “brain stunt.” A child’s brain builds a million new neural connections every second, so the wrong input can cause the brain to become miswired, she explains.
Technical errors in AI-generated educational content can create dangerous feedback loops. AI educator and creator Carla Engelbrecht says these mixed signals slow a child’s ability to learn cause and effect. This takes executive functions offline for nonsensical processing, such as cognitive delays.
Systemic risk and criminal risk
Beyond its cognitive impact, AI facilitates risky forms of exploitation. “This is perhaps one of the most shocking and visible forms of harm, and it comes in the form of the sharing of AI-generated sexually abusive content and nudity. [using AI and deepfake technology to make a person appear nude] Apps, new ways to use AI to both approach and exploit children,” says Livingstone.
The United Nations notes that organizations fighting child exploitation are reporting an increase in the distribution and creation of such content, suggesting that the number of victims is increasing. Additionally, AI tools are trained on data collected from children without adequate privacy protections, creating a permanent digital footprint on individuals who cannot legally consent.
Livingstone calls for greater clarity and recognition of how the AI challenge looks from different parts of the world and different segments of the population.
Current developments suggest that without intervention, a generation may emerge with significant gaps in basic thinking skills. To reduce these risks, the United Nations proposes several actions, including:
-
Introduction of pediatric AI standards: Experts call for the involvement of educators and child experts in the design phase of LLMs to ensure that they support, rather than replace, developmental milestones.
-
Labeling of regulatory content: Platforms like YouTube are under increasing pressure to implement content authentication for animated media to distinguish between human-curated educational content and automated “slop.”
-
Parental literacy and engagement: There is a great need for a toolkit that helps parents move from passive observation to active auditing of their children’s interactions with AI.
-
education reform: Schools need to move from evaluating output to evaluating thinking processes. This ensures that the neural pathways for critical analysis are established before delegation to the AI is allowed.
Exposing children to AI at an early age poses multifaceted challenges, including cognitive development, physical safety, and data privacy. The United Nations notes that the transition from human to AI-assisted development requires a robust multidisciplinary framework that prioritizes the long-term cognitive health of pediatric populations over the short-term efficiency of algorithmic tools.
